Cosmetic ingredient database-driven cosmetic adverse reaction prediction method and system

Abstract

The present invention discloses a cosmetics adverse reaction prediction method and system driven by a cosmetics ingredient database, which relates to the field of computer technology. The method comprises: obtaining the properties and ingredient information of a target cosmetic, predicting the audience characteristics of the target cosmetic based on the information, and forming a multi-cluster audience characteristic stream. The cosmetics attribute and ingredient information are used to perform registration optimization on the cosmetics ingredient database to determine the registration ingredient library. Multiple adverse reaction prediction learners and loss parsers are used for learning to establish dual channels for immediate and delayed adverse reaction prediction. The cosmetics ingredients and audience characteristics are input into the dual channels to generate a cosmetics adverse reaction prediction report. The method solves the technical problem that the existing cosmetics adverse reaction prediction cannot provide personalized and accurate predictions. By realizing personalized and accurate predictions of cosmetics adverse reactions, the technical effect of improving prediction accuracy is achieved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a cosmetics adverse reaction prediction method and system driven by a cosmetics ingredient database. Background Art

[0002] With growing concern about cosmetic safety, predicting adverse reactions to cosmetics has become a significant issue within the cosmetics industry. Traditional prediction methods primarily rely on ingredient comparisons or general population responses. This approach fails to accurately reflect individual differences, resulting in predictions that are neither personalized nor accurate. Furthermore, existing technologies often focus solely on predicting immediate reactions, neglecting delayed reactions after long-term use, making predictions incomplete. The development of big data and artificial intelligence technologies is enabling more precise analysis based on cosmetic ingredient information and user characteristics, providing more personalized predictions and assessments. However, current technologies still struggle to achieve truly personalized and accurate predictions, failing to fully account for the diverse needs and potential risks of each user. Therefore, providing more accurate and personalized predictions for cosmetic adverse reactions remains a major challenge facing current technologies.

[0003] At present, relevant technologies have the technical problem of being unable to provide personalized and accurate predictions for adverse reactions to cosmetics. Summary of the Invention

[0004] This application solves the technical problem that existing cosmetic adverse reaction predictions cannot provide personalized and accurate predictions by providing a cosmetic ingredient database-driven cosmetic adverse reaction prediction method and system.

[0005] This application provides a cosmetics ingredient database-driven cosmetics adverse reaction prediction method, including:

[0006] Obtain a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes cosmetic attribute information and cosmetic ingredient information corresponding to a target cosmetic; perform audience feature prediction on the target cosmetic based on the cosmetic attribute information to obtain a multi-cluster audience feature stream; perform registration optimization on a cosmetic ingredient database based on the cosmetic attribute information and the cosmetic ingredient information to determine a registered cosmetic ingredient library; introduce multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library to build a cosmetic adverse reaction prediction dual channel, wherein the cosmetic adverse reaction prediction dual channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel; input the cosmetic ingredient information and the multi-cluster audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report.

[0007] This application provides a cosmetics adverse reaction prediction system driven by a cosmetics ingredient database, including:

[0008] A prediction instruction acquisition module, the prediction instruction acquisition module is used to obtain a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes cosmetic attribute information and cosmetic ingredient information corresponding to the target cosmetic; an audience feature prediction module, the audience feature prediction module is used to perform audience feature prediction on the target cosmetic based on the cosmetic attribute information, and obtain multiple clusters of audience feature streams; a database registration optimization module, the database registration optimization module is used to perform registration optimization on the cosmetic ingredient database based on the cosmetic attribute information and the cosmetic ingredient information, and determine a registered cosmetic ingredient library; a prediction dual-channel construction module, the prediction dual-channel construction module is used to introduce multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and build a cosmetic adverse reaction prediction dual channel, wherein the cosmetic adverse reaction prediction dual channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel; an adverse reaction prediction report generation module, the adverse reaction prediction report generation module is used to input the cosmetic ingredient information and the multiple clusters of audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report.

[0009] The cosmetic ingredient database-driven cosmetic adverse reaction prediction method and system proposed in this application first obtains the properties and ingredient information of the target cosmetics. Then, based on this information, the audience characteristics of the target cosmetics are predicted, forming a multi-cluster audience characteristic stream. The cosmetic ingredient database is then optimized for registration using the cosmetic attribute and ingredient information to determine the registered ingredient library. Multiple adverse reaction prediction learners and loss parsers are used for learning, establishing a dual channel for immediate and delayed adverse reaction prediction. The cosmetic ingredients and audience characteristics are input into the dual channels to generate a cosmetic adverse reaction prediction report, achieving the technical effect of improving prediction accuracy by enabling personalized and accurate prediction of cosmetic adverse reactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0011] Figure 1A schematic flow chart of a cosmetics adverse reaction prediction method driven by a cosmetics ingredient database provided in an embodiment of the present application;

[0012] Figure 2 Schematic diagram of the structure of the cosmetics adverse reaction prediction system driven by the cosmetics ingredient database provided in the embodiment of the present application.

[0013] Explanation of the accompanying symbols: prediction instruction acquisition module 10, audience feature prediction module 20, database registration optimization module 30, prediction dual-channel construction module 40, adverse reaction prediction report generation module 50. DETAILED DESCRIPTION

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0015] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0016] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0017] The present application embodiment provides a cosmetics ingredient database driven cosmetics adverse reaction prediction method, such as Figure 1 As shown, the method includes:

[0018] Step S100, obtain a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes the cosmetic attribute information and cosmetic ingredient information corresponding to the target cosmetic. Specifically, obtain a cosmetic adverse reaction prediction instruction, which contains the attribute information and ingredient information of the target cosmetic, providing basic data for subsequent analysis. Cosmetic attribute information includes product category, function, applicable skin type, brand, usage method, etc., which is mainly used to match similar products in the database and predict the target user group to improve the accuracy of personalized adverse reaction assessment. Cosmetic ingredient information covers specific ingredient names, ingredient categories, content, risk levels and functional descriptions. By matching with the cosmetic ingredient database, the system can search for similar products, calculate ingredient similarity, and evaluate whether the cosmetic contains known high-risk ingredients, such as sensitizers or irritating preservatives. To ensure data consistency, the system standardizes the cosmetic attribute and ingredient information, including unified naming, matching CAS numbers, concentration normalization and risk level labeling, thereby constructing a structured input data set. Finally, the dataset is stored in a database and can be called through an API interface or a machine learning model to provide precise support for subsequent audience characteristic predictions, ingredient optimization matching, and adverse reaction analysis, thereby improving the reliability and scientific nature of the prediction system.

[0019] Step S200, predicting the audience characteristics of the target cosmetics based on the cosmetics attribute information, and obtaining multiple clusters of audience characteristic streams. Specifically, the target cosmetics' attribute information is used to identify its potential user groups, and the audience characteristics are segmented so that subsequent adverse reaction predictions can be personalized for different groups of people. First, based on product category, function, applicable skin type and other information, the system matches historical data and queries market user portraits to determine the target audience pool. Subsequently, core characteristics such as age, gender, and skin type are collected from the target audience, and combined with data such as regional environment, lifestyle and skin allergy history, a complete audience characteristic library is constructed. Next, the system classifies the target audience by age-gender and skin type. For example, women aged 18-25 prefer light moisturizing products, people with oily skin prefer oil-control cosmetics, and users with sensitive skin are easily irritated by certain ingredients. Based on the classification results, the system constructs a multi-cluster audience feature stream, converting the characteristics of different population groups into structured data streams, such as "Audience Cluster 1: 18-25 years old, female, oily skin" and "Audience Cluster 2: 26-35 years old, female, dry skin." This stream can be used as data input for parallel computing, improving the predictive system's adaptability and accuracy for different groups. Ultimately, this multi-cluster audience feature stream provides data support for subsequent adverse reaction predictions, enabling the model to conduct customized analysis based on different population characteristics, improving the scientific nature and reliability of the predictions.

[0020] In one possible implementation, audience characteristics are predicted for the target cosmetics based on the cosmetics attribute information to obtain multiple audience characteristic streams. Step S200 further includes step S210, where audience characteristics are predicted for the target cosmetics based on the cosmetics attribute information to determine the target audience. Specifically, audience prediction determines the target audience of the target cosmetics by analyzing the cosmetics attribute information. First, the system predicts the potential user group of the cosmetics based on the product's category, functional characteristics, and applicable skin type. For example, an oil-control foundation is suitable for young people with oily skin, while an anti-aging serum is suitable for women over 40. Next, the system collects characteristic information such as the target audience's age, gender, and skin type to construct a target audience characteristic library. The system further subdivides the groups by age, gender, and skin type classification, such as 18-25-year-old women with oily skin and 30-40-year-old men with sensitive skin. Finally, the system integrates the classification results into multiple audience characteristic streams, providing accurate data support for subsequent personalized adverse reaction predictions, thereby improving the accuracy and targeted nature of the predictions.

[0021] Step S220, collects the age information, gender information and skin quality information of the target audience to obtain a target audience feature library. Specifically, by collecting the age, gender and skin quality information of the target audience, a target audience feature library is constructed. First, the system predicts the target audience based on the attribute information of the cosmetics (such as product category, function, applicable skin quality, etc.), and collects the age, gender and skin quality data of the audience through questionnaires, e-commerce platform analysis and social media data. With respect to age information, the system divides the audience into different age groups (such as 18-25 years old, 26-35 years old, etc.), and further refines the audience classification according to gender and skin quality (such as oily, dry, mixed, sensitive skin). By integrating the above data, the system establishes a database containing audience characteristics to ensure that subsequent adverse reaction predictions and personalized recommendations can be analyzed based on accurate user data. The above process provides important support for the safety assessment of cosmetics and improves the accuracy of product matching and adverse reaction prediction.

[0022] Step S230, age and gender classification is performed according to the target audience feature library to obtain multiple clusters of audience initial classification results. Specifically, based on the target audience feature library, the system first performs age and gender classification to determine the target audience group. The system predicts the audience through the attribute information of the target cosmetics (such as product category, efficacy and applicable skin type). Taking age as the dimension, the system divides the audience into multiple age groups, such as young people aged 18-25, who usually need oil control, moisturizing, and basic skin care products, such as oil-control foundation, moisturizing spray, etc.; young women aged 26-35 are more concerned about whitening and anti-aging effects, and are suitable for using whitening essences containing niacinamide and vitamin C; while women aged 36-45 begin to pay attention to anti-aging and firming skin care products, such as anti-aging essences, firming creams, etc.; people over 46 years old usually have loose and dry skin, and tend to use moisturizing and anti-aging repair products. Next, the system further refines the classification based on gender information. Women typically have more diverse needs, ranging from basic skincare to beauty products and anti-aging serums. Men, on the other hand, tend to prefer basic care products like oil control, cleansing, and sunscreen. For example, men typically prefer simple cleansers and moisturizers, while women opt for products with fragrance or more cosmetic effects. Using this method, the system divides the target audience into multiple groups. For example, "18-25 years old, female, oily skin" may tend to choose refreshing moisturizing and oil control products, while "36-45 years old, male, sensitive skin" may prefer gentle, restorative skincare. Finally, the system categorizes these segmented audiences into multiple initial audience clusters, each with similar characteristics and needs, facilitating subsequent personalized recommendations and adverse reaction prediction. This classification not only helps identify the potential market for cosmetics but also provides accurate data support for subsequent precise predictions and safety assessments.

[0023] In step S240, skin type classification is performed based on the initial classification results of the multiple audience clusters to obtain the multiple audience feature streams. Specifically, the system segments the target audience based on each audience's skin type information (e.g., oily, dry, combination, sensitive), ensuring that different skin types receive the most appropriate skincare recommendations. For example, audiences with oily skin often face issues such as shine, enlarged pores, and acne, and therefore tend to choose oil-controlling, refreshing skincare products, such as oil-control toners, lightweight foundations, and refreshing creams. The system groups these groups into a cluster, such as "Women aged 18-25, oily skin," whose needs focus on oil control and moisturizing. On the other hand, audiences with dry skin, whose skin lacks moisture and is prone to feeling tight and dry, typically require more nourishing and hydrating products, such as creams and serums containing ingredients such as hyaluronic acid and ceramides. The system groups the user into the "Women aged 26-35, dry skin" cluster and prioritizes highly moisturizing products when recommending products. For people with combination skin, the T-zone is usually oily and the U-zone is dry. This group prefers products that balance oil and moisture, such as lightweight moisturizing lotions and oil-control essences. For example, for the "36-45 year old male with combination skin" group, balanced products are recommended to address skin problems in different areas. Finally, people with sensitive skin react violently to external stimuli, easily causing redness, swelling, and allergies. Therefore, they need gentle, non-irritating skin care products, such as repairing lotions and creams with low-sensitivity ingredients. This type of user is classified as the "45+ year old male with sensitive skin" cluster, and recommendations focus on gentle and highly restorative products. Through skin type classification, the system segments the audience into different clusters, provides personalized product recommendations for each group, and generates multi-cluster audience feature streams, providing an accurate basis for subsequent adverse reaction predictions, ensuring that the skin care needs of each group are met, while also improving the accuracy of cosmetic adverse reaction predictions.

[0024] In step S300, the cosmetic ingredient database is aligned and optimized based on the cosmetic attribute information and the cosmetic ingredient information to determine the aligned cosmetic ingredient database. Specifically, the cosmetic ingredient database alignment and optimization is performed by analyzing the attribute information and ingredient information of the target cosmetic to optimize and align the existing cosmetic ingredient database, thereby ensuring the accuracy of the ingredient data. First, the system screens the ingredient database based on the cosmetic attribute information, such as product category (such as sunscreen, anti-aging essence, face cream, etc.), efficacy (such as whitening, anti-aging, moisturizing, etc.), and applicable skin types (such as dry skin, oily skin, sensitive skin, etc.). For example, for a moisturizing face cream, the system first screens out ingredients with moisturizing functions, such as hyaluronic acid, glycerin, ceramide, etc. Then, based on the cosmetic ingredient information, the system evaluates the twin degree of each ingredient with the target cosmetic ingredient, that is, calculates the similarity of their structure, chemical properties, and efficacy. For example, hyaluronic acid and glycerin generally have a high similarity in moisturizing effect, so their alignment coefficient is high, making them suitable for use in moisturizing products. Next, the system sets an ingredient registration threshold and filters out ingredients with a registration coefficient greater than the threshold. For example, if the target cosmetic is an anti-aging essence, the system may set the registration coefficient to 0.8, and only ingredients with a registration coefficient greater than 0.8, such as retinol and peptides, will be selected into the final formula. Finally, the system generates a registered cosmetic ingredient library, which contains all optimized and screened ingredients, which not only have a high degree of matching, but also meet the functional requirements of the product. For example, for an anti-aging essence, the registered ingredient library may include hyaluronic acid, retinol and peptide ingredients, which have significant effects in anti-aging and skin repair. The registered cosmetic ingredient library will serve as an important reference for subsequent adverse reaction prediction, personalized recommendations and product development, and provide data support to ensure the safety and effectiveness of cosmetics.

[0025] In one possible implementation, a cosmetic ingredient database is aligned and optimized based on the cosmetic attribute information and the cosmetic ingredient information to determine a aligned cosmetic ingredient library. Step S300 further includes step S310, where the cosmetic ingredient database is associated and screened based on the cosmetic attribute information to obtain a cosmetic ingredient library with similar attributes. Specifically, the cosmetic ingredient database is associated and screened based on the attribute information of the target cosmetic to select a set of ingredients that match the target cosmetic's function, category, and applicable skin type. For example, if the target cosmetic is a moisturizing cream, the system will screen the ingredient database for ingredients with moisturizing properties, such as hyaluronic acid, glycerin, and ceramides. These ingredients generally have good moisture-binding properties and are suitable for dry and sensitive skin. If the target cosmetic is an anti-aging serum, the system will screen for ingredients with anti-aging properties, such as retinol and peptides, which help stimulate skin regeneration and reduce fine lines and wrinkles. Furthermore, if the target cosmetic is a sunscreen, ingredients with UV protection, such as titanium dioxide, zinc oxide, and avobenzone, will be screened for ingredients that effectively block UV rays and reduce the risk of skin aging and sunburn. During the screening process, the system further refines the selection based on the target product's applicable skin type, ensuring that the ingredients meet the needs of different skin types. For example, oily skin is suitable for oil-control ingredients such as salicylic acid and tea tree oil, while dry skin requires more moisturizing ingredients such as squalene and glycerin. Ultimately, the system screens out all ingredients that meet the target cosmetic's attribute requirements, forming a library of cosmetic ingredients with similar attributes, which contains all ingredients suitable for the target product. These ingredients will provide a foundation for subsequent formula optimization and adverse reaction prediction, ensuring the functionality and safety of the target cosmetic.

[0026] In step S320, twinning evaluation is performed on each ingredient in the same-attribute cosmetic ingredient library based on the cosmetic ingredient information to obtain multiple ingredient registration coefficients. Specifically, twinning evaluation is performed on each ingredient in the same-attribute cosmetic ingredient library using the ingredient information of the target cosmetic to calculate a registration coefficient for each ingredient. This involves a comprehensive assessment of each ingredient's chemical structure, functionality, and skin reactivity. For example, in a moisturizing cream, hyaluronic acid, a common moisturizing ingredient, is often evaluated as a high match due to its simple structure and effective moisturizing properties, potentially resulting in a higher registration coefficient, such as 0.9. Glycerin, also a moisturizing ingredient, while having a good moisturizing effect, may be less effective in certain high-efficiency formulas, resulting in a registration coefficient of 0.85. Furthermore, niacinamide, an ingredient with whitening, anti-aging, and anti-inflammatory properties, while having a good moisturizing effect, is not primarily moisturizing, unlike hyaluronic acid, and therefore may receive a lower registration coefficient, such as 0.7. Based on the matching coefficients of these ingredients, the system selects the ingredients that best meet the target product requirements and prioritizes those with high matching coefficients for inclusion in the final formula. For example, in this moisturizing cream, hyaluronic acid and glycerin might be selected, while niacinamide might be excluded. In this way, the system ensures a close match between the selected ingredients in terms of function and effect, while providing a scientific basis for subsequent formulation optimization and adverse reaction prediction, ensuring product efficacy and safety.

[0027] Step S330, the plurality of ingredient alignment coefficients are optimized and selected according to the ingredient alignment threshold to obtain an ingredient alignment optimization distribution that is greater than or equal to the ingredient alignment threshold. Specifically, the alignment coefficient of each ingredient is first screened according to the set ingredient alignment threshold to ensure that only those ingredients that are highly matched with the target cosmetics are selected. For example, when developing an anti-aging essence, the functional requirements of the target product include anti-aging and skin repair, so the system sets a alignment threshold of 0.8. Assume that after twin degree evaluation, the alignment coefficient of hyaluronic acid is 0.9, the alignment coefficient of retinol is 0.85, the alignment coefficient of glycerin is 0.75, and the alignment coefficient of fragrance ingredients is 0.3. The system will add the alignment coefficients of hyaluronic acid and retinol to the ingredient alignment optimization distribution because their alignment coefficients are greater than the set threshold of 0.8, while glycerin and fragrance ingredients will be excluded because their alignment coefficients are lower than the threshold. Ultimately, the generated optimal distribution of ingredients for alignment includes hyaluronic acid and retinol, with alignment coefficients of 0.9 and 0.85, respectively. This indicates that they closely match the functional requirements of the target anti-aging serum, providing the desired moisturizing and anti-aging effects. In this way, the system effectively selects the ingredients that best meet the needs of the target cosmetic product and provides strong support for product formulation optimization.

[0028] Step S340: Filter the homologous cosmetic ingredient library based on the component alignment optimization distribution to generate the aligned cosmetic ingredient library. Specifically, the alignment coefficient of each ingredient is first screened based on the component alignment threshold to ensure that the selected ingredient is highly compatible with the functional requirements of the target cosmetic product. Specifically, when developing an anti-aging serum, the system sets a alignment threshold of 0.8 based on the requirements of the target product. For example, after an initial twin-degree evaluation, the system determines that the alignment coefficients for hyaluronic acid are 0.9, retinol is 0.85, glycerin is 0.75, and fragrance ingredients are 0.3. Since the alignment coefficients of hyaluronic acid and retinol are both greater than 0.8, meeting the moisturizing and anti-aging functional requirements of the target product, they are retained, while glycerin and fragrance ingredients with alignment coefficients below 0.8 are excluded. Finally, the system adds the alignment coefficients of hyaluronic acid and retinol to the component alignment optimization distribution, forming an optimized library containing highly matching ingredients. Through this screening process, the system ensures that the final registered cosmetic ingredient library contains only those ingredients that are highly compatible with the target product. These ingredients, such as hyaluronic acid and retinol, can not only effectively provide anti-aging and moisturizing effects, but also ensure the efficacy and safety of the product.

[0029] Step S400, introduce multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and build a dual channel for cosmetic adverse reaction prediction, wherein the dual channel for cosmetic adverse reaction prediction includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel. Specifically, by introducing multiple adverse reaction prediction learners and adverse reaction prediction loss parsers, the target cosmetic ingredient library is subjected to adverse reaction prediction loss optimization learning, and a dual channel for cosmetic adverse reaction prediction is built. The dual channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel. For example, for an anti-aging essence, the system first analyzes its ingredients such as retinol and fragrance ingredients through the immediate adverse reaction prediction channel to predict whether the above ingredients will cause immediate reactions such as skin tingling and allergies in a short period of time. If the system finds through historical data that the fragrance ingredient has a higher probability of allergic reaction, it will immediately give a warning and may recommend reducing the proportion of the fragrance ingredient. The system also uses a delayed adverse reaction prediction channel to assess the effects of these ingredients over long-term use. For example, long-term use of retinol may cause skin dryness or irritation. Therefore, the system predicts this in the delayed channel, reminding users to monitor their skin condition or adjust their usage frequency. By combining these two channels, the system can predict the safety of cosmetics during use in real time and over the long term, providing guidance for formula optimization and adjustment, ensuring that the final product is both effective and safe on the market.

[0030] In one possible implementation, multiple adverse reaction prediction learners and adverse reaction prediction loss parsers are introduced to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and a dual channel for cosmetic adverse reaction prediction is established, wherein the dual channel for cosmetic adverse reaction prediction includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, and step S400 further includes step S410, performing adverse reaction record retrieval based on the registered cosmetic ingredient library to obtain an immediate adverse reaction record library and a delayed adverse reaction record library. Specifically, an adverse reaction record retrieval is first performed based on the registered cosmetic ingredient library to obtain immediate and delayed adverse reaction records for each ingredient. First, the system establishes an immediate adverse reaction record library based on the historical data of the ingredients, consumer feedback, and clinical trial data, recording the reactions that the ingredients may cause in the short term (such as a few minutes to a few hours). For example, fragrance ingredients are often reported to cause immediate reactions such as skin allergies, redness, and stinging. All of these data will be collected and stored in the immediate adverse reaction record library. The system then establishes a delayed adverse reaction database based on long-term usage data and product trial records. This database records potential reactions to cosmetic ingredients that may occur with long-term use, such as dry skin, hyperpigmentation, or peeling. For example, while retinol is effective for anti-aging, long-term use may cause dryness or peeling, which would be recorded in the delayed adverse reaction database. By comprehensively searching for both immediate and delayed adverse reactions to these ingredients, the system can better understand the potential risks of each ingredient at different stages of use, providing a scientific basis for product formulation optimization and safety assessment.

[0031] Step S420, audience characteristics are collected according to the immediate adverse reaction record library to obtain the first audience characteristics record library. Specifically, the relevant audience characteristics are first extracted from the immediate adverse reaction record library, and the first audience characteristics record library is established by collecting information such as the user's age, gender, skin quality, allergy history, etc. For example, assuming that a skin care product containing fragrance ingredients produces an allergic reaction in a group of 20-30-year-old females with oily skin, the system will extract the characteristic information of the group (age, gender, skin quality, reaction type) and store it in the first audience characteristics record library. Similarly, if some users experience immediate reactions such as dry skin or stinging when using an anti-aging essence containing retinol, the system will record the above-mentioned immediate reactions and associate the data with the corresponding audience characteristics (such as dry skin, allergy history, etc. in women aged 30-40). For each audience group, the system will also collect information on skin type (such as oily, dry, sensitive, etc.) and whether there is a history of allergies, so that subsequent predictions are more personalized. Through the above features, the system can establish an accurate first record library of audience characteristics based on the type of immediate adverse reactions and the specific information of the audience, providing strong data support for subsequent cosmetic formula optimization and adverse reaction prediction.

[0032] In step S430, audience characteristics are collected based on the delayed adverse reaction record library to obtain a second audience characteristic record library. Specifically, audience characteristics associated with long-term use are extracted from the delayed adverse reaction record library and stored in the second audience characteristic record library. For example, if an anti-aging serum containing retinol causes dryness and peeling after long-term use, the system will record the delayed reaction data and associate it with the age, gender, skin type, and other characteristics of the specific audience. Female users aged 30-40 with dry skin may experience more significant skin dryness or peeling when using retinol, so this data will be recorded as a specific audience characteristic. Similarly, for male users aged 20-30 with oily skin using the same product, the system may not record any significant delayed adverse reactions, as users with oily skin generally have a higher tolerance for retinol. The system also collects other characteristics for different audience groups, such as whether there is a history of allergies and frequency of use, especially for products containing ingredients that can cause pigmentation or allergic reactions with long-term use, such as certain preservatives or fragrances. Through the data, the system can generate a complete second record library of audience characteristics, providing a more accurate basis for subsequent risk predictions and ensuring the safety of the product in long-term use.

[0033] Step S440, according to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, the registered cosmetic ingredient library, the first record library of audience characteristics and the immediate adverse reaction record library are subjected to immediate adverse reaction prediction loss optimization learning to generate the immediate adverse reaction prediction channel. Specifically, through multiple adverse reaction prediction learners (machine learning models) and adverse reaction prediction loss parsers, the registered cosmetic ingredient library, the first record library of audience characteristics and the immediate adverse reaction record library are subjected to immediate adverse reaction prediction loss optimization learning to generate the immediate adverse reaction prediction channel. For example, the system uses a decision tree to analyze whether certain ingredients (such as spices) will cause allergic reactions in a specific audience (such as women aged 30-40). Through the training of the machine learning model, the system can identify which ingredients will cause allergies in users with oily skin and cause fewer reactions in users with dry skin. The support vector machine helps the system process the complex relationship between different ingredients and user characteristics, especially to classify the immediate reactions under the combined action of multiple characteristics (such as skin type, allergy history, age, etc.). During model training, a loss parser optimizes the learner by calculating the error between predicted results and actual reactions. For example, if a fragrance ingredient triggers an unexpected allergic reaction in certain user groups, the loss parser will adjust the model to reduce this error and readjust the ingredient's usage ratio. Through this process, multiple learners ultimately generate an immediate adverse reaction prediction pipeline that can predict immediate adverse reactions to cosmetics, such as allergies, stinging, and redness, in real time.

[0034] Step S450, according to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, the registered cosmetic ingredient library, the second record library of audience characteristics and the delayed adverse reaction record library are subjected to delayed adverse reaction prediction loss optimization learning to generate the delayed adverse reaction prediction channel. Specifically, the registered cosmetic ingredient library, the second record library of audience characteristics and the delayed adverse reaction record library are subjected to delayed adverse reaction prediction loss optimization learning by multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to generate a delayed adverse reaction prediction channel. For example, when the system analyzes ingredients such as retinol, multiple machine learning models (such as decision trees, neural networks, random forests) will be used to learn the long-term effects of retinol on users of different age groups and skin types. Assuming that for users with dry skin over 40 years old, the system may find that the retinol ingredient can cause delayed reactions such as dry skin and peeling after long-term use. After the model inputs these data, it is further analyzed by a support vector machine, and the system can identify which age groups, genders and skin types of users are most likely to experience side effects. During this process, the loss parser optimizes the model's predictions by calculating the error of each model. For example, if the system's prediction of a delayed adverse reaction that a certain ingredient may cause differs from the actual data, the loss parser will adjust based on the error to reduce this prediction bias and improve accuracy. Through this approach, the system continuously optimizes the prediction model and generates a final delayed adverse reaction prediction channel, which can provide real-time warnings of possible side effects after long-term use, such as dry skin, pigmentation, and long-term allergies, thereby providing data support for optimizing the safety of product formulas.

[0035] Step S460 connects the immediate adverse reaction prediction channel and the delayed adverse reaction prediction channel as parallel nodes to generate the cosmetic adverse reaction prediction dual channel. Specifically, the system first generates an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, and independently analyzes the adverse reactions of cosmetic ingredients in short-term and long-term use. The immediate adverse reaction prediction channel focuses on short-term reactions, such as immediate adverse reactions such as allergies, stinging or redness that fragrance ingredients may cause in users with oily skin. Through multiple machine learning models (such as decision trees and support vector machines), the system learns and predicts these short-term reactions, issues warnings in real time and provides formula adjustment suggestions. On the other hand, the delayed adverse reaction prediction channel targets adverse reactions after long-term use, such as problems such as dry skin, peeling or pigmentation that retinol ingredients may cause in long-term use. Through deep learning and neural network models, the system can predict the occurrence of these delayed side effects after several weeks or months of use, and provide timely suggestions, such as reducing the frequency of use or changing ingredients. To improve system efficiency and prediction accuracy, these two channels are connected via parallel nodes, ensuring that both immediate and delayed reactions can be processed and independently predicted simultaneously. This ultimately creates a dual-channel system for predicting cosmetic adverse reactions. This dual-channel system not only promptly identifies short-term adverse reactions, such as allergic reactions caused by fragrances, but also predicts long-term risks, such as the dryness or peeling that retinol may cause on dry skin. This allows for a comprehensive assessment of cosmetic safety and provides precise recommendations for product optimization.

[0036] In one possible implementation, according to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, the registered cosmetic ingredient library, the first record library of audience characteristics and the immediate adverse reaction record library are subjected to immediate adverse reaction prediction loss optimization learning to generate the immediate adverse reaction prediction channel. Step S440 further includes step S441, using the registered cosmetic ingredient library and the first record library of audience characteristics as input information and the immediate adverse reaction record library as output information to supervise the multiple adverse reaction prediction learners respectively. After each training for a predetermined number of training times, multiple immediate adverse reaction prediction loss coefficients are calculated according to the adverse reaction prediction loss parser. Specifically, the system uses the registered cosmetic ingredient library and the first record library of audience characteristics as input data and the immediate adverse reaction record library as output information to perform multiple training to generate an immediate adverse reaction prediction model. First, multiple adverse reaction prediction learners (such as decision trees, support vector machines, neural networks, etc.) are used to train the model. By learning the relationship between cosmetic ingredients and audience characteristics, it predicts which ingredients may cause allergies, stinging, redness, swelling and other reactions in the short term. For example, fragrance ingredients may cause allergies in users with oily skin, but have less of an effect on those with dry skin. During each training session, the system calculates an immediate adverse reaction prediction loss coefficient based on the adverse reaction prediction loss parser to assess the error between the predicted result and the actual reaction. For example, if the model predicts that a certain ingredient (such as a preservative) will cause an allergic reaction in 20% of users, but actual data shows that only 15% of users experience a reaction, the loss parser will calculate the error and adjust the model parameters. If the loss coefficient falls below the preset immediate adverse reaction prediction loss threshold, the system deems the model to have achieved the required accuracy and generates multiple immediate adverse reaction prediction models. These models are linked together using ensemble learning methods (such as weighted averaging or voting) to form an immediate adverse reaction prediction pipeline. For example, for a particular cosmetic product, the system can predict an immediate allergic reaction to fragrance ingredients in users with oily skin, while also predicting the potential tingling sensation caused by preservative ingredients in sensitive skin, ultimately providing R&D personnel with safety assessments and formula adjustment recommendations.

[0037] Step S442: If the multiple immediate adverse reaction prediction loss coefficients are less than the immediate adverse reaction prediction loss threshold, multiple immediate adverse reaction prediction models are generated. Specifically, the system uses the registered cosmetic ingredient library and the first record library of audience characteristics as input data, and the immediate adverse reaction record library as output data, and performs supervised training through multiple adverse reaction prediction learners (such as decision trees, support vector machines, neural networks, etc.). During each training, the system uses the data in the training set (including cosmetic ingredients and user characteristics) to predict the immediate reactions that cosmetic ingredients may cause to different groups (such as oily skin, dry skin or sensitive skin users), such as allergies, redness or stinging. Through training, the model will gradually optimize and identify which ingredients are more likely to cause these reactions in specific audience groups. For example, fragrance ingredients may cause stinging or allergic reactions in users with dry skin, but may not cause such reactions in users with oily skin. The system calculates the loss coefficient of each training through the adverse reaction prediction loss parser to measure the error between the predicted result and the actual reaction. If multiple immediate adverse reaction prediction loss coefficients are less than the set immediate adverse reaction prediction loss threshold, the system believes that the model has achieved the expected accuracy and will generate multiple immediate adverse reaction prediction models. For example, the system will generate a decision tree model that can identify the immediate reactions of fragrance ingredients to different age groups and skin types, and adjust its predictions based on past training data. Next, the system will connect these models through an integrated learning method (such as weighted average or voting mechanism) to form the final immediate adverse reaction prediction channel, which can predict in real time the immediate adverse reactions that may be caused by cosmetic ingredients during use, and provide accurate early warnings for product formula optimization and safety assessment.

[0038] Step S443 connects the multiple immediate adverse reaction prediction models to obtain the immediate adverse reaction prediction channel. Specifically, in the cosmetics adverse reaction prediction system, when multiple immediate adverse reaction prediction models are generated through training, the system connects the models using an ensemble learning method to form an immediate adverse reaction prediction channel. For example, assume there are three models: a decision tree, a support vector machine (SVM), and a neural network. Each model learns the relationship between cosmetic ingredients and the target audience based on different characteristics. The decision tree model may find that fragrance ingredients have a stronger allergic reaction in users with oily skin, while the SVM model may predict that fragrance ingredients are more irritating to sensitive skin. The neural network provides a comprehensive prediction based on more dimensional data (such as age, allergy history, etc.). The system integrates the model predictions through methods such as weighted averaging or voting mechanisms. For example, if both the decision tree and neural network models predict that a fragrance ingredient will cause an allergic reaction in users with dry skin, but the SVM model does not predict such a reaction, the system will make a final judgment based on the predictions of the majority model. In addition, the performance of each model during the training process determines its weight in the ensemble, with the better performing model receiving a higher weight. Ultimately, by integrating these models, the immediate adverse reaction prediction channel can comprehensively assess the immediate reactions that cosmetics may cause, such as redness, stinging, and allergies, in user groups with different skin types, ages, and allergy histories. For example, if a fragrance-containing cream is predicted to cause redness and swelling in female users with sensitive skin, the system will provide feedback, suggesting reducing the use of fragrance or adjusting the ingredient ratio, thereby optimizing the product formula and improving its safety.

[0039] In one possible implementation, multiple adverse reaction prediction learners and adverse reaction prediction loss parsers are introduced to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and a dual-channel for cosmetic adverse reaction prediction is established, wherein the dual-channel for cosmetic adverse reaction prediction includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel. Step S400 further includes step S470, and the adverse reaction prediction loss parser includes an adverse reaction prediction loss parsing function, and the adverse reaction prediction loss parsing function is: Among them, LOSS represents the bad prediction loss coefficient, M represents the scheduled number of training times, m represents the mth training, M and m are both positive integers, 1≤m≤M, SUO m Represents the number of adverse reaction prediction samples in the mth training, SUX mRepresents the number of correct adverse reaction prediction samples in the mth training. Specifically, in the cosmetics adverse reaction prediction system, the adverse reaction prediction loss parser evaluates the accuracy of the model prediction by calculating the loss coefficient. During each training process, the system calculates the loss coefficient for each round of training based on the adverse reaction prediction loss parsing function, which is in the form of: Among them, SUO m Indicates the number of adverse reaction prediction samples in the mth round of training, SUX m represents the number of correctly predicted samples, and M is the number of training rounds. For example, in a training round, if the system processes 100 samples and successfully predicts 80 correct reactions, the loss coefficient is 0.2, indicating that the model's prediction accuracy needs improvement. After each training round, the system evaluates model performance based on the calculated loss coefficient. If the loss coefficient is below a preset loss threshold, the model's prediction is sufficiently accurate. At this point, multiple models are combined to generate a final immediate adverse reaction prediction model. For example, suppose the system discovers through training that fragrance ingredients can trigger allergic reactions in users of certain age groups. After calculating the loss coefficient using the loss parser, if the loss coefficient remains above the threshold, the system adjusts model parameters, such as reducing the fragrance dosage or optimizing the formula for sensitive groups, until the loss coefficient meets the required level. The resulting prediction model will accurately reflect the immediate reactions of fragrances to users with different skin types, ages, and allergy histories. This process continuously optimizes the loss coefficient to ensure that each prediction accurately identifies potential adverse reactions in cosmetics, providing a scientific basis for product safety.

[0040] Step S500: Input the cosmetic ingredient information and the multi-cluster audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report. Specifically, in the cosmetic adverse reaction prediction system, the system uses the cosmetic ingredient information and the multi-cluster audience feature streams as input data and transmits them to the cosmetic adverse reaction prediction dual channel for processing. First, the system predicts short-term (immediate adverse reactions) and long-term (delayed adverse reactions) based on the cosmetic ingredient information, such as hyaluronic acid, retinol, preservatives, and fragrances, combined with the multi-cluster audience feature streams (including the user's skin type, age, gender, allergy history, etc.). For example, the system will analyze the immediate reaction of fragrance ingredients to users with oily skin and predict possible allergic reactions; at the same time, it will also evaluate the impact of retinol ingredients after long-term use, especially the risk of dry skin or peeling in users with dry skin. The immediate adverse reaction prediction channel will process reactions such as redness, allergies, or stinging that may be caused by fragrance ingredients, and make different predictions for different skin types and age groups. The delayed adverse reaction prediction channel analyzes the long-term use effects of cosmetic ingredients to predict problems such as skin dryness and pigmentation that may be caused by long-term use of retinol. Ultimately, the system combines the prediction results of these two channels to generate a detailed cosmetic adverse reaction prediction report, which lists the short-term and long-term reactions of each ingredient in different user groups. For example, fragrance ingredients may cause allergies in female users with sensitive skin, while retinol may cause dryness or peeling in users with dry skin over 40 years old. Through this dual-channel processing, R&D personnel can adjust the product formula according to the report, such as reducing the concentration of fragrances or adjusting the content of retinol, thereby optimizing the safety of the product and ensuring its applicability in the market.

[0041] In one possible implementation, the cosmetic ingredient information and the multi-cluster audience feature streams are input into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report. Step S500 further includes step S510, in which the cosmetic ingredient information and the multi-cluster audience feature streams are input into the immediate adverse reaction prediction channel to obtain a multi-cluster audience-immediate adverse reaction prediction result. Specifically, in the cosmetic adverse reaction prediction system, the cosmetic ingredient information and the multi-cluster audience feature streams are used as input and enter the immediate adverse reaction prediction channel for processing. First, the system analyzes cosmetic ingredient information, such as fragrances, retinol, preservatives, and other ingredients, and combines them with the multi-cluster audience feature streams, which include information such as the user's age, gender, skin type (such as oily skin, dry skin, sensitive skin), and allergy history. The system uses multiple machine learning models (such as decision trees, SVMs, neural networks, etc.) to evaluate the immediate reactions of these ingredients in different user groups. For example, fragrance ingredients may cause allergic reactions in oily skin groups, but almost no reaction in dry skin users. The system trains models to accurately predict the immediate adverse reactions of different ingredients to each group, such as redness, stinging, and allergies. If the decision tree model predicts that the probability of allergic reactions to fragrance ingredients in oily skin groups is 80%, while the neural network model predicts that the probability of redness and swelling reactions caused by the same ingredient in sensitive skin users is 60%, the system will generate immediate adverse reaction prediction results for each audience group based on the prediction results of these different models. These results will help the R&D team identify the immediate allergic reactions that cosmetic ingredients may cause in different user groups, providing a scientific basis for product formula adjustments. Ultimately, the multi-cluster audience-immediate adverse reaction prediction results output by the system will list in detail the immediate reaction risks that specific cosmetic ingredients may cause in each audience group, thereby providing data support for product optimization and safety assessment.

[0042] Step S520, the cosmetic ingredient information and the multi-cluster audience feature stream are input into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse reaction prediction result. Specifically, the cosmetic ingredient information and the multi-cluster audience feature stream are input into the delayed adverse reaction prediction channel, the main task of which is to evaluate the potential impact of cosmetic ingredients on different groups after long-term use. The system first analyzes the information of cosmetic ingredients, such as retinol, hyaluronic acid, preservatives, etc., which may have an impact on the skin after a few weeks or months. For example, retinol may cause peeling or dryness for users with dry skin, but has less impact on oily skin groups. Then, the multi-cluster audience feature stream provides detailed information for each user group, including age, gender, skin type (such as oily, dry, sensitive skin), allergy history, etc. Based on this information, the system uses multiple machine learning models (such as decision trees, SVM, neural networks, etc.) to predict the potential reactions of different ingredients in long-term use. For example, for dry skin groups over 40 years old, the system may predict that the probability of retinol ingredients causing dry skin or peeling is 60%, while for young oily skin groups, this probability may be only 20%. Based on the characteristics of each group, the system will calculate the delayed reaction risk of each cosmetic ingredient in that group, such as pigmentation, dry skin or allergies. Ultimately, the system outputs multi-cluster audience-delayed adverse reaction prediction results, providing personalized delayed adverse reaction predictions for each group, helping the R&D team evaluate the safety of ingredients and optimize product formulas. For example, if the system detects that retinol may cause peeling reactions in dry skin groups, the R&D team may choose to reduce the concentration of the ingredient or develop a retinol-free alternative product for such users.

[0043] Step S530 performs data fusion based on the multiple clusters of audience-based immediate adverse reaction prediction results and the multiple clusters of audience-based delayed adverse reaction prediction results, and outputs the cosmetic adverse reaction prediction report. Specifically, the cosmetic adverse reaction prediction report, generated by data fusion of the immediate adverse reaction prediction results and the delayed adverse reaction prediction results, evaluates each ingredient's potential adverse reactions in the short term (immediate reaction) and long term (delayed reaction), based on the characteristics of different audience groups (such as skin type, age, gender, allergy history, etc.) and cosmetic ingredient information. For example, fragrance ingredients may trigger a higher rate of immediate allergic reactions in users with oily skin, while retinol ingredients may cause dryness and peeling after long-term use in users with dry skin. The immediate adverse reaction prediction channel and the delayed adverse reaction prediction channel predict these reactions separately, generating their own results. The system then combines these two prediction results using a weighted average or voting mechanism. For example, the immediate reaction prediction probability of fragrance ingredients in users with oily skin is predicted to be 80%, while the delayed reaction prediction probability in users with dry skin is predicted to be 20%. The system uses a weighted calculation to assign different weights based on group characteristics to produce a comprehensive assessment. If fragrance ingredients have a higher risk of immediate reactions in people with sensitive skin, while retinol has a higher risk of delayed reactions in people with dry skin, the report will issue warnings for each of these ingredients, prompting the R&D team to reduce the fragrance content or adjust the retinol concentration. Furthermore, the system comprehensively considers the characteristics of each group through a decision tree approach. The resulting report will detail the immediate and delayed reaction risks for each group and provide recommendations for formula optimization, such as recommending reducing fragrance use for users with sensitive skin or providing a retinol-free alternative product version for those with dry skin.

[0044] The embodiment of the present application obtains the properties and ingredient information of the target cosmetics, and then predicts the audience characteristics of the target cosmetics based on this information, forming a multi-cluster audience characteristic stream. The cosmetics attribute and ingredient information are used to perform registration optimization on the cosmetics ingredient database to determine the registration ingredient library. Multiple adverse reaction prediction learners and loss parsers are used for learning, and a dual channel for immediate and delayed adverse reaction prediction is established. The cosmetic ingredients and audience characteristics are input into the dual channel to generate a cosmetics adverse reaction prediction report, achieving the technical effect of improving prediction accuracy by realizing personalized and accurate prediction of cosmetics adverse reactions.

[0045] In the above, refer to Figure 1 The cosmetics adverse reaction prediction method driven by the cosmetics ingredient database according to the embodiment of the present invention is described in detail. Figure 2 The following describes a cosmetics adverse reaction prediction system driven by a cosmetics ingredient database according to an embodiment of the present invention.

[0046] The cosmetics ingredient database-driven cosmetics adverse reaction prediction system according to an embodiment of the present invention addresses the technical issue of existing cosmetics adverse reaction prediction systems being unable to provide personalized and accurate predictions. By enabling personalized and accurate predictions of cosmetics adverse reactions, the system achieves the technical effect of improving prediction accuracy. The cosmetics ingredient database-driven cosmetics adverse reaction prediction system includes a prediction instruction acquisition module 10, an audience feature prediction module 20, a database alignment optimization module 30, a dual-channel prediction establishment module 40, and an adverse reaction prediction report generation module 50.

[0047] The prediction instruction acquisition module 10 is used to obtain a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes cosmetic property information and cosmetic ingredient information corresponding to the target cosmetic.

[0048] The audience feature prediction module 20 is used to predict the audience features of the target cosmetics according to the cosmetics attribute information, and obtain multiple clusters of audience feature streams.

[0049] The database registration and optimization module 30 is used to perform registration and optimization on the cosmetic ingredient database according to the cosmetic attribute information and the cosmetic ingredient information, and determine the registered cosmetic ingredient database.

[0050] The prediction dual-channel building module 40 is used to introduce multiple adverse reaction prediction learners and adverse reaction prediction loss analyzers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and build a cosmetic adverse reaction prediction dual channel, wherein the cosmetic adverse reaction prediction dual channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel.

[0051] The adverse reaction prediction report generating module 50 is used to input the cosmetic ingredient information and the multi-cluster audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report.

[0052] The specific configuration of the audience feature prediction module 20 will be described in detail below. As described above, the audience feature prediction of the target cosmetics is performed based on the cosmetics attribute information to obtain multiple clusters of audience feature streams. The audience feature prediction module 20 further includes: a target audience population determination unit, the target audience population determination unit is used to perform audience prediction for the target cosmetics based on the cosmetics attribute information to determine the target audience population; a target audience feature library acquisition unit, the target audience feature library acquisition unit is used to collect the age information, gender information and skin quality information of the target audience population to obtain a target audience feature library; an initial classification result acquisition unit, the initial classification result acquisition unit is used to perform age and gender classification based on the target audience feature library to obtain multiple clusters of audience initial classification results; a skin quality classification unit, the skin quality classification unit is used to perform skin quality classification based on the multiple clusters of audience initial classification results to obtain the multiple clusters of audience feature streams.

[0053] The specific configuration of the database registration optimization module 30 will be described in detail below. As described above, the cosmetic ingredient database is registered and optimized according to the cosmetic attribute information and the cosmetic ingredient information to determine the registered cosmetic ingredient library. The database registration optimization module 30 further includes: a cosmetic ingredient library acquisition unit, the cosmetic ingredient library acquisition unit is used to perform association screening on the cosmetic ingredient database according to the cosmetic attribute information to obtain a cosmetic ingredient library with the same attribute; a twin degree evaluation unit, the twin degree evaluation unit is used to perform twin degree evaluation on each cosmetic ingredient data with the same attribute in the cosmetic ingredient library with the same attribute according to the cosmetic ingredient information to obtain multiple ingredient calibration coefficients; an optimization selection unit, the optimization selection unit is used to optimize and select the multiple ingredient calibration coefficients according to the ingredient calibration threshold to obtain a component calibration optimization distribution greater than or equal to the ingredient calibration threshold; a cosmetic ingredient library screening unit, the cosmetic ingredient library screening unit is used to screen the cosmetic ingredient library with the same attribute according to the component calibration optimization distribution to generate the registered cosmetic ingredient library.

[0054] The specific configuration of the prediction dual-channel building module 40 will be described in detail below. As described above, multiple adverse reaction prediction learners and adverse reaction prediction loss parsers are introduced to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and a cosmetic adverse reaction prediction dual channel is built, wherein the cosmetic adverse reaction prediction dual channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, and the prediction dual channel building module 40 further includes: an adverse reaction record library acquisition unit, the adverse reaction record library acquisition unit is used to retrieve adverse reaction records according to the registered cosmetic ingredient library, and obtain an immediate adverse reaction record library and a delayed adverse reaction record library; an audience feature first record library acquisition unit, the audience feature first record library acquisition unit is used to collect audience features according to the immediate adverse reaction record library, and obtain an audience feature first record library; an audience feature second record library acquisition unit, the audience feature second record library acquisition unit is used to collect audience features according to the delayed adverse reaction record library, and obtain an audience feature second record library; an immediate adverse reaction record library. An adverse reaction prediction channel generation unit, wherein the immediate adverse reaction prediction channel generation unit is used to perform immediate adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, the first audience feature record library, and the immediate adverse reaction record library based on the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser to generate the immediate adverse reaction prediction channel; a delayed adverse reaction prediction channel generation unit, wherein the delayed adverse reaction prediction channel generation unit is used to perform delayed adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, the second audience feature record library, and the delayed adverse reaction record library based on the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser to generate the delayed adverse reaction prediction channel; a parallel node connection unit, wherein the parallel node connection unit is used to connect the immediate adverse reaction prediction channel and the delayed adverse reaction prediction channel as parallel nodes to generate the cosmetics adverse reaction prediction dual channel.

[0055] Wherein, according to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, the registered cosmetic ingredient library, the first record library of audience characteristics and the immediate adverse reaction record library are subjected to immediate adverse reaction prediction loss optimization learning to generate the immediate adverse reaction prediction channel, and the immediate adverse reaction prediction channel generation unit further includes: a loss coefficient calculation subunit, the loss coefficient calculation subunit is used to use the registered cosmetic ingredient library and the first record library of audience characteristics as input information and the immediate adverse reaction record library as output information, and respectively perform supervised training on the multiple adverse reaction prediction learners, and calculate multiple immediate adverse reaction prediction loss coefficients according to the adverse reaction prediction loss parser every time a predetermined number of trainings are trained; an adverse reaction prediction model generation subunit, the adverse reaction prediction model generation subunit is used to generate multiple immediate adverse reaction prediction models if the multiple immediate adverse reaction prediction loss coefficients are less than the immediate adverse reaction prediction loss threshold; an immediate adverse reaction prediction channel acquisition subunit, the immediate adverse reaction prediction channel acquisition subunit is used to connect the multiple immediate adverse reaction prediction models to obtain the immediate adverse reaction prediction channel.

[0056] The prediction dual-channel building module 40 further includes: an adverse reaction prediction loss parser component unit, wherein the adverse reaction prediction loss parser component unit is used for the adverse reaction prediction loss parser to include an adverse reaction prediction loss parsing function, and the adverse reaction prediction loss parsing function is: Among them, LOSS represents the bad prediction loss coefficient, M represents the scheduled number of training times, m represents the mth training, M and m are both positive integers, 1≤m≤M, SUO m Represents the number of adverse reaction prediction samples in the mth training, SUX m Represents the number of correct samples for adverse reaction prediction in the mth training.

[0057] The specific configuration of the adverse reaction prediction report generation module 50 will be described in detail below. As described above, the cosmetic ingredient information and the multi-cluster audience feature stream are input into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report. The adverse reaction prediction report generation module 50 further includes: a multi-cluster audience-immediate adverse prediction result acquisition unit, the multi-cluster audience-immediate adverse prediction result acquisition unit is used to input the cosmetic ingredient information and the multi-cluster audience feature stream into the immediate adverse reaction prediction channel to obtain a multi-cluster audience-immediate adverse prediction result; a multi-cluster audience-delayed adverse prediction result acquisition unit, the multi-cluster audience-delayed adverse prediction result acquisition unit is used to input the cosmetic ingredient information and the multi-cluster audience feature stream into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse prediction result; a data fusion unit, the data fusion unit is used to perform data fusion based on the multi-cluster audience-immediate adverse prediction result and the multi-cluster audience-delayed adverse prediction result, and output the cosmetic adverse reaction prediction report.

[0058] The cosmetic ingredient database-driven cosmetic adverse reaction prediction system provided in the embodiment of the present invention can execute the cosmetic ingredient database-driven cosmetic adverse reaction prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0059] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0060] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. A cosmetics adverse reaction prediction method driven by a cosmetics ingredient database, characterized in that: The method comprises: Obtaining a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes cosmetic attribute information and cosmetic ingredient information corresponding to the target cosmetic; Predicting audience characteristics of the target cosmetics based on the cosmetics attribute information to obtain multiple clusters of audience characteristic streams; Performing registration and optimization on the cosmetic ingredient database according to the cosmetic attribute information and the cosmetic ingredient information to determine the registered cosmetic ingredient database includes: Performing association screening on the cosmetic ingredient database according to the cosmetic attribute information to obtain a cosmetic ingredient database with the same attribute; Performing twin degree evaluation on each cosmetic ingredient data of the same attribute in the cosmetic ingredient library according to the cosmetic ingredient information to obtain a plurality of ingredient calibration coefficients; Optimizing and selecting the plurality of component calibration coefficients according to a component calibration threshold to obtain a component calibration optimal distribution that is greater than or equal to the component calibration threshold; Screening the cosmetic ingredient library with the same attribute according to the component registration optimal distribution to generate the registered cosmetic ingredient library; Introducing multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and building a dual channel for cosmetic adverse reaction prediction, wherein the dual channel for cosmetic adverse reaction prediction includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel; The cosmetic ingredient information and the multi-cluster audience feature streams are input into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report.

2. The method according to claim 1, wherein Predicting audience characteristics of the target cosmetics based on the cosmetics attribute information to obtain multiple clusters of audience characteristic streams includes: Performing audience prediction for the target cosmetics based on the cosmetics attribute information to determine the target audience; Collecting age information, gender information, and skin quality information of the target audience to obtain a target audience feature database; Perform age and gender classification based on the target audience feature library to obtain multiple audience cluster initial classification results; Skin type classification is performed based on the initial classification results of the multiple audience clusters to obtain the multiple audience cluster feature streams.

3. The method according to claim 1, wherein Introducing multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and building a dual channel for cosmetic adverse reaction prediction, including: Perform adverse reaction record retrieval based on the registered cosmetic ingredient database to obtain an immediate adverse reaction record database and a delayed adverse reaction record database; Collecting audience characteristics based on the immediate adverse reaction record database to obtain a first audience characteristics record database; Collecting audience characteristics based on the delayed adverse reaction record database to obtain a second audience characteristics record database; According to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, performing immediate adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, the first audience feature record library, and the immediate adverse reaction record library to generate the immediate adverse reaction prediction channel; According to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, performing delayed adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, the second audience feature record library, and the delayed adverse reaction record library to generate the delayed adverse reaction prediction channel; The immediate adverse reaction prediction channel and the delayed adverse reaction prediction channel are connected as parallel nodes to generate the cosmetic adverse reaction prediction dual channel.

4. The method according to claim 3, wherein According to the multiple adverse reaction prediction learners and the adverse reaction prediction loss parser, performing immediate adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, the first audience feature record library, and the immediate adverse reaction record library to generate the immediate adverse reaction prediction channel, including: Using the registered cosmetic ingredient library and the first audience feature record library as input information and the immediate adverse reaction record library as output information, respectively, supervised training is performed on the multiple adverse reaction prediction learners, and each time a predetermined number of training times are trained, multiple immediate adverse reaction prediction loss coefficients are calculated according to the adverse reaction prediction loss parser; If the multiple immediate adverse reaction prediction loss coefficients are less than the immediate adverse reaction prediction loss threshold, generating multiple immediate adverse reaction prediction models; The multiple immediate adverse reaction prediction models are connected to obtain the immediate adverse reaction prediction channel.

5. The method according to claim 1, wherein The adverse reaction prediction loss parser includes an adverse reaction prediction loss parsing function, and the adverse reaction prediction loss parsing function is: ； Among them, LOSS represents the adverse reaction prediction loss coefficient, M represents the scheduled number of training times, m represents the m-th training, M and m are both positive integers, 1≤m≤M, SUOm represents the number of adverse reaction prediction samples in the m-th training, and SUXm represents the number of correct adverse reaction prediction samples in the m-th training.

6. The method according to claim 1, wherein Inputting the cosmetic ingredient information and the multi-cluster audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report, including: Inputting the cosmetic ingredient information and the multi-cluster audience feature streams into the immediate adverse reaction prediction channel to obtain a multi-cluster audience-immediate adverse reaction prediction result; Inputting the cosmetic ingredient information and the multi-cluster audience feature streams into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse reaction prediction result; Data fusion is performed based on the multiple-cluster audience-immediate adverse effect prediction results and the multiple-cluster audience-delayed adverse effect prediction results, and the cosmetic adverse reaction prediction report is output.

7. A cosmetics adverse reaction prediction system driven by a cosmetics ingredient database, characterized in that: The system is used to implement the cosmetic ingredient database-driven cosmetic adverse reaction prediction method according to any one of claims 1 to 6, and the system comprises: A prediction instruction acquisition module, the prediction instruction acquisition module is used to obtain a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction includes cosmetic attribute information and cosmetic ingredient information corresponding to the target cosmetic; An audience feature prediction module, configured to predict audience features of the target cosmetics based on the cosmetics attribute information to obtain multiple clusters of audience feature streams; A database registration and optimization module, configured to perform registration and optimization on the cosmetic ingredient database according to the cosmetic attribute information and the cosmetic ingredient information, and determine a registered cosmetic ingredient database; A prediction dual-channel building module, wherein the prediction dual-channel building module is used to introduce multiple adverse reaction prediction learners and adverse reaction prediction loss parsers to perform adverse reaction prediction loss optimization learning on the registered cosmetic ingredient library, and build a cosmetic adverse reaction prediction dual-channel, wherein the cosmetic adverse reaction prediction dual-channel includes an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel; An adverse reaction prediction report generation module is used to input the cosmetic ingredient information and the multi-cluster audience feature streams into the cosmetic adverse reaction prediction dual channel to generate a cosmetic adverse reaction prediction report.

Images