Method for establishing sweetness-increasing fragrance base module based on machine learning

Abstract

The invention relates to a method for establishing a sweetening base module based on machine learning, which comprises the following steps: selecting a sweetening function formula, constructing a sensory evaluation method for the sweetening function of a spice formula, and obtaining and verifying data of a sweetening core spice raw material and an auxiliary spice raw material; selecting a representative sweetening functional formula, and by taking the representative sweetening functional formula as a reference, imitating a sweetening core aroma raw material and an imitated functional formula by utilizing data of the sweetening core aroma raw material and the auxiliary aroma raw material; and inputting the component content and sweet sense data of the representative sweetening function formula and the imitation function formula into a pre-trained machine learning model to obtain a created formula. The invention provides a method for creating a sweetness-increasing flavor base module based on machine learning, a foundation is laid for functional analysis and digital research of tobacco flavor components, and a new thought is provided for tobacco flavor raw material flavoring control and cigarette product quality improvement.

Description

Technical Field

[0001] The present application belongs to the technical field of cigarette product design, and specifically, relates to a method for establishing a sweetening aroma base module based on machine learning. Background Art

[0002] As an important raw material in the production of cigarettes, the formulation research and development of tobacco flavors involves the core technology of the tobacco industry. Most of the current research on the creation of flavor base modules still relies mainly on sensory evaluation and empirical rules, which is labor-intensive, time-consuming and labor-intensive. However, the raw materials of tobacco flavor base modules are of many types and complex ingredients. The relationship between many raw materials and sensory functions is unclear, making it difficult to achieve targeted flavor enhancement and characteristic creation.

[0003] Therefore, it is an urgent problem for technical personnel in this field to clarify the relationship between tobacco flavor base module raw materials and sensory functions, develop new methods for creating flavor base modules, and provide new research ideas for the digital research of tobacco flavor raw materials and the improvement of cigarette product quality. Summary of the invention

[0004] This application is based on at least one of the above-mentioned technical problems, and proposes a method for establishing a sweetening flavor base module based on machine learning. The idea is to clarify the sweetening core flavor raw materials and auxiliary flavor raw materials in the formula through sensory evaluation, formula data analysis, imitation verification and machine learning model construction, and to use machine learning to construct a model through sweetness score values ​​and the content of monomer flavors in the formula. The machine learning model outputs the created formula and performs sensory verification. This invention lays the foundation for the functional analysis and digital research of tobacco flavor formula ingredients, and provides new ideas for the control of flavoring tobacco flavor raw materials and the improvement of cigarette product quality.

[0005] To achieve the above objectives, the present application proposes a method for establishing a sweetening flavor base module based on machine learning.

[0006] Select sweetening function formulas and construct sensory evaluation methods for sweetening function of flavor formulas, obtain data on sweetening core flavor raw materials and auxiliary flavor raw materials and verify them;

[0007] Select representative sweetening functional formulas and use them as a benchmark, and use the data of sweetening core aroma raw materials and auxiliary aroma raw materials to imitate the formula containing sweetening core aroma raw materials and imitation functional formulas;

[0008] The ingredient content and sweetness data of the representative sweetening functional formula and the imitation functional formula are input into a pre-trained machine learning model to obtain a creative formula.

[0009] In some embodiments, the method for establishing a pre-trained machine learning model comprises the following steps:

[0010] The sweetening function fitting output value is Y and the content of different monomer components is X. The model of calculating the sweetening function output value Y and X is fitted by machine learning.

[0011] The machine learning algorithm is one or more of the random forest algorithm, K nearest neighbor regression algorithm, support vector machine algorithm, decision tree algorithm, etc. 2 Greater than 0.80.

[0012] In some embodiments, a method for constructing a sensory evaluation method for the sweetening function of a flavor formula is used to select a flavor formula with sweetening function, and a sensory evaluation of the sweetening function is performed and the scores are summarized.

[0013] In some embodiments, constructing a sensory evaluation method for the sweetening function of a flavor formula comprises:

[0014] The nasal sweetness evaluation score and the aftertaste sweetness evaluation score are set separately; the sum of the nasal sweetness evaluation score and the aftertaste sweetness evaluation score is taken as the total score; the sweetness score is the difference between the sweetness scores of the test cigarette and the blank cigarette;

[0015] The total score and sweetness score were used to construct sweet sensory evaluation data.

[0016] In some embodiments, the similarities in the composition of formula raw materials whose sweetness scores reach a set value are grouped and the core aroma raw material monomers and auxiliary aroma raw material monomers are clearly identified, and the core aroma raw material monomers and auxiliary aroma raw material monomers are evaluated and verified separately using a sensory evaluation method for the sweetening function of a fragrance formula.

[0017] In some embodiments, 50-200 representative sweetening functional formulas are selected.

[0018] In some embodiments, there are 10-50 representative sweetening functional formulas, 1-15 sweetening core flavor raw material monomers and 0-10 auxiliary flavor raw material monomers.

[0019] In some embodiments, the monomer contents of 1-5 representative sweetening function formulas are selected as the basis, 10-90 imitation formulas containing core flavor raw materials and auxiliary flavor raw materials are configured, and the sweet sensory evaluation data are obtained by constructing a flavor formula sweetening function sensory evaluation method.

[0020] In some embodiments, the number of created formulas is 1-5, and a sensory evaluation of the sweetening function of the spice formula is performed, and the machine learning model is optimized using the sensory evaluation data of the sweetening function of the spice formula.

[0021] In this application, representative sweetening functional formulas are selected to construct a sensory evaluation method for the sweetening function of spice formulas. The core sweetening flavor raw materials and auxiliary flavor raw materials are screened out and verified through data analysis technology, and the formulas with better sweetening effects are imitated to prepare formulas containing core flavor raw materials and auxiliary flavor raw materials. The representative sweetening functional formulas and imitated functional formulas are based on the ingredient content and sweetness data, and a model is established through training of machine learning algorithms such as neural networks. The imitated formula is output according to the machine learning model, and verification and model optimization are performed. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a random forest model fitting provided in one embodiment of the present application;

[0023] Figure 2 A schematic diagram of support vector machine model fitting provided in one embodiment of the present application;

[0024] Figure 3 A schematic diagram of model fitting of a K nearest neighbor regression model provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of model fitting of a decision tree model provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0028] The invention discloses a method for creating a sweetening flavor base module based on machine learning, selects a representative sweetening functional formula, constructs a sensory evaluation method for the sweetening function of a flavor formula, selects the sweetening core flavor raw materials and auxiliary flavor raw materials through data analysis technology, and verifies them, and imitates the formula containing the core flavor raw materials and auxiliary flavor raw materials based on the formula with better sweetening effect. The representative sweetening functional formula and the imitation functional formula are based on the ingredient content and sweetness data, and a model is established through training of machine learning algorithms such as neural networks, and the imitation formula is output according to the machine learning model, and verification and model optimization are performed. The specific method includes the following steps:

[0029] Step 1: Construct a sensory evaluation method for the sweetening function of flavor formulas, select 50-200 flavor formulas with sweetening function, determine the sweetness score of each formula and summarize and analyze them; the constructed sensory evaluation method for the sweetening function of flavor formulas includes nasal sweetness (50 points) and aftertaste sweetness (50 points), and the total score is the sum of the nasal sweetness and the aftertaste sweetness; add the sweetening function formula to blank cigarettes to form test cigarettes, and 8 or more professionally trained evaluators conduct sensory evaluation on the test cigarettes according to the above sensory evaluation method, and the final sweetness score is the difference between the sweetness scores of the test cigarettes and the blank cigarettes.

[0030] Step 2: Select 10-50 formulas with higher sweetening function scores, group them according to the similarity of the formula raw material composition, conduct raw material formula data analysis, identify the core aroma raw material monomers and auxiliary aroma raw material monomers that appear frequently in each group of formulas, and analyze 1-15 core aroma raw material monomers and 0-10 auxiliary aroma raw material monomers that appear frequently in each group of formulas, and conduct sensory evaluation and verification of the sweetening function respectively.

[0031] Step 3: Based on the monomer content of the selected formulas 1-5 with better sweetening effects, a simulated formula containing core aroma raw materials and auxiliary aroma raw materials is prepared, and the sweetening function is evaluated to obtain sweetness sensory evaluation data.

[0032] In this embodiment, 10 to 90 imitation formulas containing core aroma raw materials and auxiliary aroma raw materials are configured, and the sweetening function is evaluated to obtain sweetness sensory evaluation data.

[0033] Step 4: Select the imitation formula in step 3 and the formula with a higher sweetening function score in step 2, and use the sweetening function fitting output value as Y and the different monomer component contents as X to calculate the model of the sweetening function output value Y and X through machine learning fitting.

[0034] For example, step 4 selects 10-90 imitation formulas from step 3 and 10-50 formulas with higher sweetening function scores from step 2, with the sweetening function fitting output value being Y, and different monomer component contents being X1, X2, X3..., and a model of the sweetening function output value Y and X1, X2, X3... is calculated through machine learning fitting.

[0035] Furthermore, the machine learning algorithm can be one or more of the random forest algorithm, K nearest neighbor regression algorithm, support vector machine algorithm, decision tree algorithm, etc., requiring the R 2 Greater than 0.80.

[0036] Step 5: Based on the required sweetening score, use the machine learning model to output the creation formula, conduct sensory evaluation verification, and optimize and improve the model. That is, use the machine learning model to output 1-5 creation formulas, conduct sweetening function evaluation, and use their sweetness scores to optimize the model.

[0037] Example 1 Construction of sensory evaluation method for sweetening function and sensory evaluation of formula

[0038] According to the sensory evaluation requirements of "YC / T 497-2014 Sensory Evaluation Method for Chinese Cigarette Style" and "YC / T 496-2014 Sensory Comfort Evaluation Method for Cigarette", the sensory evaluation method was adjusted, and the sweetness score was divided into nasal sweetness (0-50 points) and aftertaste sweetness (0-50 points). The total sweetness score was the sum of the two scores (Table 1). With reference to the sweetness score gradient of common commercially available finished cigarettes (Table 2, the score of the blank cigarette was determined, and a sensory evaluation was conducted on 150 flavor formulas, and their sweetness data were statistically analyzed.

[0039] Table 1 Sensory evaluation method for sweetening function

[0040]

[0041] Table 2 Summary of sweetness scores of commercially available finished cigarettes

[0042]

[0043]

[0044] Example 2 Screening of formulas with better sweetening function

[0045] The sweetness data of 150 flavor formulas were analyzed and counted, and 30 formulas with better sweetening function were selected from them (PF1-PF30 (Table 3). The final sweetness score Y is the difference between the sweetness scores of the test cigarette and the blank cigarette. The raw material composition of the 30 groups of formulas was analyzed, and the ones with the highest frequency in each group of formulas were identified as core flavor raw material monomers and auxiliary flavor raw material monomers. Standard products were purchased and added to blank cigarettes for sensory evaluation verification. Finally, the five core flavor raw material monomers were determined to be ethyl maltol, maltol, furanone, damascenone, and MCP, and the six auxiliary flavor raw material monomers were sorbitol, imported apricot extract, sweet orange oil, Maillard reaction product A, sour plum extract, and red date extract.

[0046] Table 3 Summary of scores of 30 recipes with better sweetening function

[0047]

[0048]

[0049] Example 3: Formulation and sensory evaluation based on core fragrance raw materials and auxiliary fragrance raw materials monomers

[0050] According to the common concentration ranges of the above five core aroma raw materials and six auxiliary aroma raw materials in the five formulas, 60 imitation formulas (PF31-PF90) were designed and evaluated for sweetness scores (Table 4). Determine the sweetness scores of the 60 formulas.

[0051] Table 4 Summary of sweetness scores of 90 recipes

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] Example 4 Machine Learning Model Construction

[0061] The data of the 30 groups of formulas in Example 2 and the 60 groups of formulas in Example 3 were collected and sorted (Table 4). Taking the sweetness score as the output value Y, the contents of the 5 core aroma raw materials and the 6 auxiliary aroma raw materials in the formula were X (X1, X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X12, X13, X14, X15, X16, X17, X18, X19, X20, X21, X22, X23, X24, X25, X26, X27, X28, X29, X30, X31, X32, X33, X34, X35, X36, X37, X38, X39, X40, X41, X42, X43, X44, X45, X46, X47, X48 10 , X 11 , using machine learning python programming code, with X and Y feature data as the data source, divided into training set and validation set in a 7:3 manner, selecting random forest model, support vector machine model, K nearest neighbor regression model, decision tree model in machine learning, using specific parameters, training set for training and fitting, validation set for validation model, using mean absolute error (MAE) and determination coefficient R 2 As an evaluation, a machine learning model of sweetness score and ingredient content was established. The results are as follows Figure 1-Figure 4 The results show that the random forest model R 2 Score 0.8687, MAE = 3.3394, support vector machine model R 2 Score 0.8036, MAE = 4.5933, K nearest neighbor regression model R 2Score 0.8658, MAE = 3.9815, decision tree model R 2 The score is 0.8043 and MAE is 3.9259. The best score is the random forest model.

[0062] Example 5 Sensory Evaluation Verification of Model Output Formula

[0063] Assuming that the sweetness score of the required recipe is 35, the random forest model is used to output two recipes with the following ratios (X1 = 0.015, X3 = 0.05, X4 = 0.04, X 10 =0.01, X 11 =0.001 and (X1 = 0.02, X3 = 0.02, X6 = 0.02, X7 = 0.01, X8 = 0.01, X 10 =0.001. Two recipes were prepared according to the above ratios and injected into blank cigarettes for sensory evaluation. The results showed that the average sweetness score of the first recipe was 34, and the average sweetness score of the second recipe was 35. The actual sweetness score was very close to the theoretical prediction value, indicating that the machine learning method model is relatively accurate and reliable.

[0064] In summary, this application proposes a method for establishing a sweetening flavor base module based on machine learning. The idea is to clarify the sweetening core flavor raw materials and auxiliary flavor raw materials in the formula through sensory evaluation, formula data analysis, imitation verification, and machine learning model construction methods, and to use the sweetness score and the content of monomer flavors in the formula to build a machine learning model, and to output the machine learning model to create a formula and perform sensory verification. This method lays the foundation for the functional analysis and digital research of tobacco flavor formula ingredients, and also provides new ideas for the control of flavoring tobacco flavor raw materials and the improvement of cigarette product quality.

[0065] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for establishing a sweetening flavor base module based on machine learning, characterized in that: Select sweetening function formulas and construct sensory evaluation methods for sweetening function of flavor formulas, obtain data on sweetening core flavor raw materials and auxiliary flavor raw materials and verify them; Select representative sweetening functional formulas and use them as a benchmark, and use the data of sweetening core aroma raw materials and auxiliary aroma raw materials to imitate the formula containing sweetening core aroma raw materials and imitation functional formulas; The ingredient content and sweetness data of the representative sweetening functional formula and the imitation functional formula are input into a pre-trained machine learning model to obtain a creative formula.

2. The method according to claim 1, characterized in that The method for establishing the pre-trained machine learning model comprises the following steps: The sweetening function fitting output value is Y and the content of different monomer components is X. The model of calculating the sweetening function output value Y and X is fitted by machine learning. The machine learning algorithm is one or more of the random forest algorithm, K nearest neighbor regression algorithm, support vector machine algorithm, decision tree algorithm, etc. 2 Greater than 0.

80.

3. The method according to claim 1, characterized in that By constructing a sensory evaluation method for the sweetening function of spice formulas, spice formulas with sweetening function were selected, and the sensory evaluation of their sweetening function was carried out and the scores were summarized.

4. The method according to claim 3, characterized in that The sensory evaluation method for sweetening function of spice formula includes: The nasal sweetness evaluation score and the aftertaste sweetness evaluation score are set separately; the sum of the nasal sweetness evaluation score and the aftertaste sweetness evaluation score is taken as the total score; the sweetness score is the difference between the sweetness scores of the test cigarette and the blank cigarette; The total score and sweetness score were used to construct sweet sensory evaluation data.

5. The method according to claim 3, characterized in that: The similarities in the composition of formula raw materials whose sweetness scores reach the set value are grouped and the core aroma raw material monomers and auxiliary aroma raw material monomers are clarified. The core aroma raw material monomers and auxiliary aroma raw material monomers are evaluated and verified separately by constructing a sensory evaluation method for the sweetening function of a flavor formula.

6. The method according to claim 3, characterized in that 50-200 representative sweetening functional formulas were selected.

7. The method according to claim 3, characterized in that There are 10-50 representative sweetening functional formulas, 1-15 sweetening core flavor raw material monomers and 0-10 auxiliary flavor raw material monomers.

8. The method according to claim 3, characterized in that Based on the monomer content of 1-5 representative sweetening functional formulas, 10-90 imitation formulas containing core aroma raw materials and auxiliary aroma raw materials were configured, and the sweet sensory evaluation data were obtained by constructing a sensory evaluation method for the sweetening function of fragrance formulas.

9. The method according to claim 3, characterized in that: The number of created formulas is 1 to 5, and a sensory evaluation of the sweetening function of the spice formula is performed, and the machine learning model is optimized using the sensory evaluation data of the sweetening function of the spice formula.

Images