How to choose cosmetic containers
A data-driven method using classification and machine learning predicts suitable cosmetic containers for novel formulations, addressing the challenge of unskilled selection by providing high-similarity results for easy container choice.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- POLA CHEMICAL INDUSTRIES INC
- Filing Date
- 2024-11-08
- Publication Date
- 2026-05-20
AI Technical Summary
Unskilled individuals face difficulty in selecting a suitable container for novel cosmetic formulations due to a lack of judgment skills, making it challenging to choose the appropriate container type.
A method involving data classification, similarity calculation, and machine learning to predict the physical properties of new cosmetic formulations, enabling the selection of suitable containers based on high similarity with existing formulations.
Enables unskilled users to easily select suitable cosmetic containers by referencing high-similarity results, reducing the need for actual measurements and computation time, and simplifying the selection process.
Smart Images
Figure 2026083901000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for selecting a container for cosmetics for selecting a container suitable for a formulation of a novel cosmetic.
Background Art
[0002] Conventionally, as containers for cosmetics, there are tube types, bottle types, jar types, pump types, etc. (for example, Patent Document 1). When manufacturing a novel cosmetic, a method of selecting a container of a type suitable for the formulation of the novel cosmetic is adopted based on the judgment of a skilled person.
Prior Art Documents
Patent Documents
[0003]
Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] According to the above conventional method, although a skilled person can easily select a container of a type suitable for the formulation of a novel cosmetic, those with little selection work experience other than skilled persons and those without experience (hereinafter referred to as "unskilled persons") cannot appropriately judge which container is suitable for the formulation of the novel cosmetic, and there is a problem that it is difficult to select a container.
[0005] The present invention has been made to solve the above problems, and an object thereof is to provide a method for selecting a container for cosmetics that enables an unskilled person to easily select a container suitable for the formulation of a novel cosmetic.
Means for Solving the Problems
[0006] To achieve the above objective, the method for selecting a cosmetic container according to claim 1 is characterized in that a processing unit performs the following steps: a storage step of associating and storing classified data, which is data classified for each combination of a plurality of container types and a plurality of existing formulation types for each of a plurality of existing cosmetic products, with the formulation and physical properties of the classified data; a new formulation acquisition step of acquiring a new formulation, which is a formulation of a new cosmetic product; a physical properties acquisition step of acquiring the physical properties of the new cosmetic product; a predetermined combination acquisition step of acquiring a predetermined combination, which is a combination of a predetermined container type, which is one of a plurality of container types, and a predetermined existing formulation type, which is one of a plurality of existing formulation types; a similarity calculation step of calculating the similarity between the new formulation and the classified data formulation of the predetermined combination; a selection step of selecting classified data up to a predetermined rank in order of the highest similarity to the new formulation from the similarity calculation results in the similarity calculation step; and an output step of outputting the new cosmetic product and the physical properties of the classified data up to the predetermined rank.
[0007] According to this cosmetic container selection method, a new formulation is obtained, the physical properties of the new cosmetic are obtained, a predetermined combination is obtained which is a combination of a predetermined container type (one of several container types) and a predetermined existing formulation type (one of several existing formulation types), and the similarity between the new formulation and the classified formulation data of the predetermined combination is calculated. Then, the classified data up to a predetermined rank is selected from the similarity calculation results in order of the highest similarity to the new formulation, and the physical properties of the new cosmetic and the classified data are output. As a result, an inexperienced person can refer to the output results to check whether there are existing cosmetics with similar physical properties that have a high similarity to the formulation of the new cosmetic, among the classified data of combinations of the predetermined container type and predetermined existing type. If such a combination exists, the inexperienced person can recognize that the predetermined container type is suitable for the container of the new cosmetic and can easily select a suitable container for the new cosmetic.
[0008] In the present invention, in the physical property acquisition step, it is preferable that predicted values of the physical properties of the new cosmetic are obtained using a machine learning model.
[0009] According to this method for selecting cosmetic containers, the physical properties of a new formulation are predicted using a machine learning model, eliminating the need to actually measure the physical properties of the new formulation. This reduces the burden on the user when selecting a container.
[0010] In the present invention, it is preferable that in the similarity calculation step, one of the following is calculated as the similarity: Euclidean distance, Manhattan distance, Mahalanobis distance, Minkowski distance, and cosine similarity.
[0011] Here, Euclidean distance, Manhattan distance, Mahalanobis distance, Minkowski distance, and cosine similarity are commonly used as indicators of similarity between data in hierarchical cluster analysis and other methods. Therefore, according to this method for selecting cosmetic containers, the similarity between new formulations and classified data can be appropriately calculated using such common indicators.
[0012] In the present invention, in the similarity calculation step, it is preferable to perform dimensionality reduction on the classification data of the new formulation and the predetermined combinations before calculating the similarity.
[0013] According to this method for selecting cosmetic containers, the similarity is calculated after dimensionality reduction of the classified data formulations for new formulations and predetermined combinations. This reduces the computation time and computation load when calculating the similarity between new formulations and classified data formulations for predetermined combinations. [Brief explanation of the drawing]
[0014] [Figure 1] This figure shows an information processing device for performing a method for selecting a cosmetic container according to one embodiment of the present invention. [Figure 2]This figure shows the classification structure of the classified data. [Figure 3] This is a flowchart showing the processing of classified data. [Figure 4] This flowchart shows the processing of highly similar data. [Figure 5] This is a diagram showing the input screen. [Figure 6] This is a diagram showing the type combination selection screen. [Figure 7] This is a diagram showing a screen displaying highly similar data. [Modes for carrying out the invention]
[0015] The following describes a method for selecting a cosmetic container according to one embodiment of the present invention, with reference to the drawings. As will be described later, the container selection method of this embodiment is for selecting a container suitable for a new cosmetic formulation.
[0016] The specific processing in the selection method of this embodiment is performed by the information processing device 1 shown in Figure 1. This information processing device 1 is of the personal computer type and includes a display 1a, a main unit 1b, and an input interface 1c, etc. The main unit 1b includes storage such as an HDD, a processor, and memory (RAM, E2PROM, ROM, etc.) (none of which are shown).
[0017] The storage of the main unit 1b of this device has application software installed for performing various processes, which will be described later. The input interface 1c consists of a keyboard and mouse for operating the information processing device 1.
[0018] Next, the classified data of the cosmetics used in this embodiment will be described. The classified data used in this embodiment classifies the data of 603 formulations of existing cosmetics (hereinafter referred to as "existing formulation data") into five types of container types based on the similarity of containers, and in the existing formulation data of each container type, it is classified into seven types of formulation types based on the similarity of formulations.
[0019] Specifically, as shown in FIG. 2, in the classified data, the container type is classified into five types: bottle, vial, pump type, tube, and others, and at the same time, the formulation type is classified into seven types: types 1 to 7.
[0020] Next, the classified data processing will be described while referring to FIG. 3. In this processing, each formulation and physical property value of the classified data classified as described above is for storing in the storage of the apparatus main body 1b. In the following description, the operation of the input interface 1c by the user is referred to as "user operation".
[0021] In this classified data processing, first, the input processing of the classified data is executed (FIG. 3 / STEP1). In this input processing, the formulations and physical property values (for example, pH and viscosity, etc.) in each of the classified data classified as described above are input by user operation.
[0022] Next, the storage processing of the classified data is executed (FIG. 3 / STEP2). In this storage processing, the formulations and physical property values of the classified data input as described above are stored in the information processing apparatus 1 in a state where they are associated with each other. In this embodiment, the storage processing corresponds to the storage step.
[0023] Next, the high-similarity data processing will be described while referring to FIG. 4. This processing is to obtain data with a high degree of formulation similarity for a new formulation from the classified data described above and display it on the display 1a, and is executed by the information processing apparatus 1 at a predetermined control cycle.
[0024] As shown in the figure, first, it is determined whether or not the execution conditions for this process are met (Figure 4 / STEP10). In this case, if any of the following conditions (a1) to (a2) are met, it is determined that the execution conditions for this process are met; otherwise, it is determined that the execution conditions for this process are not met.
[0025] (a1) The start operation of this process was executed for the first time after the information processing device 1 was started up. (a2) If the termination operation of this process was executed at a previous control timing, the start operation of this process was executed at a control timing after the termination operation.
[0026] In this case, the start of this process is initiated by the user pressing a start button (not shown), and the end of this process is initiated by the user pressing an end button 12b (see Figure 7), which will be described later.
[0027] If this determination is negative (Figure 4 / STEP10...NO) and the execution conditions for this process are not met, the process ends. On the other hand, if this determination is positive (Figure 4 / STEP10...YES) and the execution conditions for this process are met, it is determined whether or not a new prescription has already been entered (Figure 4 / STEP11). If this determination is positive (Figure 4 / STEP11...YES) and a new prescription has already been entered, the process proceeds to STEP15, which will be described later.
[0028] On the other hand, if this determination is negative (Figure 4 / STEP11...NO) and no new prescription has been entered, the input screen display process is executed (Figure 4 / STEP12). In this input screen display process, the input screen 10 shown in Figure 5 is displayed on the display 1a. This input screen 10 is for entering data for a new prescription.
[0029] As shown in Figure 5, the text "Please enter the data for the new prescription. After entering, please press the execute button" is displayed at the top of the input screen 10, and below it, there are input fields 10a for numerous prescription data such as "Pure water (wt%)" (only three are shown). Furthermore, an execute button 10b is displayed at the bottom right of the input screen 10.
[0030] Next, it is determined whether or not a new prescription has been entered (Figure 4 / STEP 13). In this case, if the user has entered each data item of the new prescription into the input field 10a and the execute button 10b is pressed, it is determined that a new prescription has been entered; otherwise, it is determined that no new prescription has been entered. In this embodiment, this determination process corresponds to the new prescription acquisition step.
[0031] If this determination is negative (Figure 4 / STEP13...NO) and no new prescription has been entered, this process ends. On the other hand, if this determination is positive (Figure 4 / STEP13...YES) and a new prescription has been entered, the predicted value calculation process is executed (Figure 4 / STEP14).
[0032] In this prediction value calculation process, although a detailed explanation is omitted, the predicted physical properties of the new formulation are calculated using the method proposed by the applicant in Japanese Patent Application No. 2020-126447, using data on the new formulation and a machine learning model (such as Random Forest and XGBoost). In this embodiment, the prediction value calculation process corresponds to the physical property acquisition step.
[0033] Next, it is determined whether a type combination has already been selected (Figure 4 / STEP15). If this determination is positive (Figure 4 / STEP15...YES) and a type combination has already been selected, the process proceeds to STEP20, which will be described later. On the other hand, if this determination is negative (Figure 4 / STEP15...NO) and a type combination has not yet been selected, the process of displaying the type combination selection screen is executed (Figure 4 / STEP16).
[0034] In this type combination selection screen display process, the type combination selection screen 11 shown in Figure 6 is displayed. This type combination selection screen 11 is for selecting a combination of each of the five container types mentioned above and each of the seven formulation types (Type 1 to Type 7) mentioned above.
[0035] As shown in Figure 6, the text "Please press the button for the container type and formulation type combination. Then, press the execute button." is displayed at the top of the type combination selection screen 11, and below it, several type combination buttons 11a such as "Bottle - Type 1" are displayed (only three are shown). Furthermore, an execute button 11b is displayed at the bottom right of the combination selection screen 11.
[0036] Next, it is determined whether a type combination has been selected (Figure 4 / STEP 17). In this case, if the user presses one of the many type combination buttons 11a and then the execute button 11b is pressed, it is determined that a type combination has been selected; otherwise, it is determined that no type combination has been selected. In this embodiment, this determination process corresponds to the predetermined combination acquisition step.
[0037] If this determination is negative (Figure 4 / STEP17...NO) and no type combination is selected, the process ends. On the other hand, if this determination is positive (Figure 4 / STEP17...YES) and a type combination is selected, the similarity processing is performed (Figure 4 / STEP18). In the following explanation, we will use the case where the "bottle-type 1" combination is selected as the type combination as an example.
[0038] In this similarity calculation process, first, the dimensions (dimensions of the ingredient data) of each prescription and new prescription for the selected "bottle-type 1" combinations are reduced using a predetermined dimensionality reduction algorithm. In this case, UMAP (Uniform Manifold Approximation and Projection) or an Auto-encoder is used as the predetermined dimensionality reduction algorithm.
[0039] Next, the similarity between each of the classified data of the "bottle-type 1" combinations, whose dimensionality has been reduced as described above, and the data of the new formulation is calculated. Specifically, the Euclidean distance is calculated as the similarity. In this embodiment, the similarity calculation process corresponds to the similarity calculation step.
[0040] Next, a high-similarity data selection process is executed (Figure 4 / STEP 19). In this high-similarity data acquisition process, classified data up to a predetermined rank (e.g., rank 5) is selected as high-similarity data, starting with those whose Euclidean distance from the new prescription data is closest to the calculation result of the similarity calculation process described above. In this embodiment, the high-similarity data selection process corresponds to the selection step.
[0041] Next, the high-similarity data display process is executed (Figure 4 / STEP20). In this high-similarity data display process, the high-similarity data screen 12 shown in Figure 7 is displayed on the display 1a. At the top of this high-similarity data screen 12, the words "To finish, press the exit button" and "Bottle - Type 1" are displayed, and below that, the data field 12a is displayed, along with the exit button 12b in the lower right corner.
[0042] This data field 12a shows the predicted physical properties of the new formulation and the physical properties of the highly similar data mentioned above. The physical properties displayed include multiple types such as "pH" and "viscosity" (only two are shown). The existing formulations 1-5 in the figure are the five formulations from the highly similar data mentioned above that have the closest Euclidean distance to the new formulation. Furthermore, the exit button 12b is pressed by the user to terminate this process. In this embodiment, the highly similar data display process corresponds to the output step.
[0043] By referring to the display results in Figure 7, users can check whether there are any existing formulations with similar physical properties among existing formulations 1-5 that have a high degree of similarity to the new formulation. If there are existing formulations with similar physical properties that have a high degree of similarity to the new formulation, users can determine whether a bottle-type container is suitable for that existing formulation by referring to the usage history of that existing formulation in a bottle-type container.
[0044] As a result, if it is determined that a bottle-type container is not suitable for the existing formulation, the process shown in Figure 4 can be repeated, and a combination other than "Bottle-Type 1" can be selected to continue the process of finding a suitable container type for the new formulation.
[0045] As described above, according to the cosmetic container selection method of this embodiment, when a new formulation and a type combination, which is a combination of one formulation type and one container type, are input, the Euclidean distance is calculated as the similarity between the new formulation and the classified data formulations in the type combination. Based on the similarity calculation result, classified data up to a predetermined rank are selected in order of the highest similarity to the new formulation, the predicted physical properties of the new formulation are calculated, and the predicted physical properties of the new formulation and the physical properties of the classified data up to the predetermined rank are displayed.
[0046] As a result, even if the user is inexperienced, they can refer to the displayed results to check whether there are any classified data with similar physical properties to the new formulation among the classified data of combinations of a given container type and a given existing type, where the similarity to the formulation is high. If such a combination exists, the inexperienced user can recognize that the given container type is suitable for the new formulation and can easily select a suitable container for the new formulation.
[0047] Furthermore, by using a machine learning model to calculate predicted physical properties of new formulations, it becomes unnecessary to actually measure the physical properties of new formulations. This reduces the burden on users when selecting containers. In addition, by calculating the Euclidean distance between new formulations and classified data formulations after dimensionality reduction, the computation time and computational load required to calculate the Euclidean distance between them can be reduced.
[0048] In this embodiment, a personal computer type processing unit 1 is used as the processing unit. However, a cloud computing system, a server, or multiple personal computers may be used instead, or a combination of a server and personal computers may be used as the processing unit.
[0049] Furthermore, although the embodiment uses Euclidean distance as the similarity measure between data, Manhattan distance, Mahalanobis distance, Minkowski distance, and cosine similarity may be used instead.
[0050] Furthermore, while the embodiment is an example in which the similarity between the new prescription and the classified data is calculated after reducing the dimensions of the new prescription and the classified data using a predetermined dimensionality reduction algorithm, the similarity between the two may also be calculated without reducing the dimensions of the new prescription and the classified data.
[0051] On the other hand, the embodiment is an example in which existing prescription data is classified into 7 types of prescriptions and 5 types of containers, but instead, the prescription types of the existing prescription data may be classified into 6 or fewer types or 8 or more types, and the container types of the existing prescription data may be classified into 4 or fewer types or 6 or more types.
[0052] Furthermore, although the embodiment is an example in which a process to display highly similar data is executed as an output step, instead, a process to print highly similar data may be executed as an output step.
[0053] Furthermore, while the embodiment uses predicted values calculated by a machine learning model as the physical properties of the new formulation, actual measured values may be used instead. [Explanation of Symbols]
[0054] 1. Information Processing Device
Claims
1. A storage step involves associating and storing classified data, which is data categorized for each of several existing cosmetic products by combinations of multiple container types and multiple existing formulation types, with the physical properties of said classified data. The new formulation acquisition step involves obtaining a new formulation, which is the formulation of a new cosmetic product. A step to obtain physical properties of the aforementioned new cosmetic product, A predetermined combination acquisition step, which acquires a predetermined combination that is a combination of a predetermined container type, which is one of the plurality of container types, and a predetermined existing formulation type, which is one of the plurality of existing formulation types, A similarity calculation step of calculating the similarity between the new formulation and the formulation of the classified data of the predetermined combination, A selection step in which, from the similarity calculation results in the similarity calculation step, the classified data are selected in order from the one with the highest similarity to the new formulation up to a predetermined rank, An output step which outputs the physical property values of the new cosmetic and the classified data up to the predetermined rank, A method for selecting a cosmetic container, characterized in that the selection process is performed by a processing unit.
2. In the method for selecting a cosmetic container according to claim 1, A method for selecting a cosmetic container, characterized in that, in the step of acquiring physical properties, predicted values of the physical properties of the new cosmetic are acquired using a machine learning model.
3. In the method for selecting a cosmetic container according to claim 1, A method for selecting a cosmetic container, characterized in that, in the similarity calculation step, one of the following is calculated as the similarity: Euclidean distance, Manhattan distance, Mahalanobis distance, Minkowski distance, and cosine similarity.
4. In the method for selecting a cosmetic container according to any one of claims 1 to 3, A method for selecting a cosmetic container, characterized in that, in the similarity calculation step, the similarity is calculated after reducing the dimensionality of the classification data of the new formulation and the predetermined combination.