CN120072116A - Cosmetic adverse reaction prediction method and system driven by cosmetic composition database - Google Patents

Abstract

The present invention discloses a cosmetics adverse reaction prediction method and system driven by a cosmetics ingredient database, and 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 according to the information, and forming a multi-cluster audience characteristic stream. The cosmetics ingredient database is registered and optimized using the cosmetics properties and ingredient information to determine the registered ingredient library. Multiple adverse reaction prediction learners and loss parsers are used for learning to build dual channels for immediate and delayed adverse reaction prediction. The cosmetic ingredients and audience characteristics are input into the dual channels to generate a cosmetics adverse reaction prediction report. The technical problem that the existing cosmetics adverse reaction prediction cannot provide personalized and accurate predictions is solved, and the technical effect of improving the prediction accuracy is achieved by realizing personalized and accurate predictions of cosmetics adverse reactions.

Description

Cosmetic adverse reaction prediction method and system driven by cosmetic composition database

Technical Field

The application relates to the technical field of computers, in particular to a method and a system for predicting adverse reactions of cosmetics driven by a cosmetic composition database.

Background

With the rise of concern about cosmetic safety, prediction of cosmetic adverse reactions has become an important issue in the cosmetic industry. The traditional prediction method mainly depends on the general response of the composition ratio to or a wide population, and the method can not accurately reflect individual differences, so that the prediction result is not personalized and accurate. Meanwhile, the prior art is only focused on the prediction of immediate response, and neglects delayed response after long-term use, so that the prediction has insufficient comprehensiveness. With the development of big data and artificial intelligence technology, more accurate analysis can be performed based on the composition information and user characteristics of cosmetics, and more personalized prediction and assessment can be provided. However, the current technology still has difficulty in achieving true individualization and accurate prediction, and cannot fully consider different needs and potential risks of each user. Therefore, how to provide more accurate and personalized prediction results in prediction of adverse reactions of cosmetics is still a major problem faced by the current technology.

In the related art at the present stage, the technical problem that personalized and accurate prediction cannot be provided in the prediction of adverse reactions of cosmetics exists.

Disclosure of Invention

The application solves the technical problem that the prior cosmetic adverse reaction prediction cannot provide personalized and accurate prediction by providing the cosmetic adverse reaction prediction method and the system driven by the cosmetic composition database.

The application provides a cosmetic adverse reaction prediction method driven by a cosmetic composition database, which comprises the following steps:

The method comprises the steps of obtaining a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction comprises cosmetic attribute information and cosmetic component information corresponding to a target cosmetic, performing audience characteristic prediction on the target cosmetic according to the cosmetic attribute information to obtain a plurality of clusters of audience characteristic flows, performing registration optimization on a cosmetic component database according to the cosmetic attribute information and the cosmetic component information to determine a registration cosmetic component library, introducing a plurality of adverse reaction prediction learners and an adverse reaction prediction loss analyzer to perform adverse reaction prediction loss optimization learning on the registration cosmetic component library, and building a cosmetic adverse reaction prediction dual-channel, wherein the cosmetic adverse reaction prediction dual-channel comprises an instant adverse reaction prediction channel and a delayed adverse reaction prediction channel, and inputting the cosmetic component information and the plurality of clusters of audience characteristic flows into the cosmetic adverse reaction prediction dual-channel to generate a cosmetic adverse reaction prediction report.

The present application provides a cosmetic adverse reaction prediction system driven by a cosmetic composition database, comprising:

The system comprises a prediction instruction acquisition module, a database registration optimizing module, a prediction dual-channel construction module and a cosmetic adverse reaction prediction generation module, wherein the prediction instruction acquisition module is used for acquiring a cosmetic adverse reaction prediction instruction, the cosmetic adverse reaction prediction instruction comprises cosmetic attribute information and cosmetic ingredient information corresponding to a target cosmetic, the audience characteristic prediction module is used for predicting audience characteristics of the target cosmetic according to the cosmetic attribute information to acquire a plurality of clusters of audience characteristic flows, the database registration optimizing module is used for carrying out registration optimizing on a cosmetic ingredient database according to the cosmetic attribute information and the cosmetic ingredient information to determine a registration cosmetic ingredient library, the prediction dual-channel construction module is used for introducing a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers to carry out adverse reaction prediction loss learning on the registration cosmetic ingredient library, the cosmetic adverse reaction prediction dual-channel is constructed, the cosmetic adverse reaction prediction dual-channel comprises an instant adverse reaction prediction channel and a delayed adverse reaction prediction channel, and the adverse reaction prediction generation module is used for inputting the predicted adverse reaction report to the cosmetic ingredient database, and the prediction dual-channel report adverse reaction prediction report is used for generating the cosmetic adverse reaction prediction report.

According to the method and the system for predicting the adverse reactions of the cosmetics driven by the cosmetic composition database, firstly, the attribute and the composition information of the target cosmetics are acquired, and then the audience characteristics of the target cosmetics are predicted according to the information, so that a multi-cluster audience characteristic stream is formed. And performing registration optimization on the cosmetic composition database by using the cosmetic attributes and the composition information, and determining a registration composition database. And learning by adopting a plurality of adverse reaction prediction learners and a loss analyzer, and constructing a dual-channel for instant and delayed adverse reaction prediction. The components of the cosmetics and the audience characteristics are input into the double channels to generate a prediction report of the adverse reaction of the cosmetics, so that the personalized and accurate prediction of the adverse reaction of the cosmetics is realized, and the technical effect of improving the prediction accuracy is achieved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly describe the drawings of the embodiments of the present disclosure, in which flowcharts are used to illustrate operations performed by a system according to embodiments of the present disclosure. It should be understood that the preceding or following operations are not necessarily performed in order precisely. Rather, the various steps may be processed in reverse order or simultaneously, as desired. Also, other operations may be added to or removed from these processes.

FIG. 1 is a flow chart of a method for predicting adverse reactions of cosmetics driven by a cosmetic composition database according to an embodiment of the present application;

fig. 2 is a schematic structural diagram of a cosmetic adverse reaction prediction system driven by a cosmetic composition database according to an embodiment of the present application.

Reference numerals illustrate the prediction instruction acquisition module 10, the audience feature prediction module 20, the database registration optimizing module 30, the prediction dual-channel construction module 40 and the adverse reaction prediction report generation module 50.

Detailed Description

The foregoing description is only an overview of the present application, and is intended to be implemented in accordance with the teachings of the present application in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present application more readily apparent.

The present application will be described in further detail below with reference to the accompanying drawings, in order to make the objects, technical solutions and advantages of the present application more apparent, and the described embodiments should not be construed as limiting the present application, but all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to 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, the term "first\second" being referred to merely as distinguishing between similar objects and not representing a particular ordering for the objects. The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements that are expressly listed or inherent to such process, method, article, or apparatus, but may include other steps or modules that may not be expressly listed or inherent to such process, method, article, or apparatus, and unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used herein is for the purpose of describing embodiments of the application only.

The embodiment of the application provides a cosmetic adverse reaction prediction method driven by a cosmetic composition database, as shown in fig. 1, comprising the following steps:

And step S100, obtaining a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction comprises cosmetic attribute information and cosmetic ingredient information corresponding to the target cosmetic. Specifically, a cosmetic adverse reaction prediction instruction is acquired, wherein the instruction contains attribute information and component information of a target cosmetic, and basic data is provided for subsequent analysis. The cosmetic attribute information comprises product types, functions, applicable skin types, brands, using modes and the like, is mainly used for matching similar products in a database, predicting target user groups and improving accuracy of personalized adverse reaction evaluation. The cosmetic ingredient information includes specific ingredient names, ingredient categories, contents, risk levels, functional descriptions, etc., and by matching with the cosmetic ingredient database, the system can find similar products, calculate ingredient similarity, and evaluate whether the cosmetic contains known high risk ingredients, such as sensitizers or stimulating preservatives. To ensure data consistency, the system performs standardized processing on cosmetic attributes and component information, including unified naming, matching CAS numbers, concentration normalization, and risk level labeling, to construct a structured input dataset. Finally, the data set is stored in a database and can be called through an API (application program interface) or a machine learning model, so that accurate support is provided for subsequent audience characteristic prediction, component optimization matching and adverse reaction analysis, and the reliability and the scientificity of a prediction system are improved.

And step S200, performing audience characteristic prediction on the target cosmetics according to the cosmetic attribute information to obtain multi-cluster audience characteristic streams. Specifically, the potential user groups of the target cosmetics are identified through the attribute information of the target cosmetics, and audience characteristics are subdivided, so that subsequent adverse reaction prediction can be subjected to personalized analysis for different groups. Firstly, the system matches historical data and queries market user portraits based on information such as product category, function, applicable skin and the like, and a target audience pool is determined. And then, core characteristics such as age, sex, skin quality and the like are collected from the target audience, and a complete audience characteristic library is constructed by combining data such as regional environment, life style, skin allergy history and the like. Next, the system classifies the target audience for age-gender and skin, such as 18-25 year old women towards light and thin moisturizing products, oily muscle groups are more prone to control oily cosmetics, and sensitive muscle users are more prone to irritation by certain components. Based on the classification result, the system builds a multi-cluster audience characteristic stream, converts the characteristics of different crowds into a structured data stream, such as 'audience cluster 1:18-25 years old, female, oily muscle', 'audience cluster 2:26-35 years old, female, xerophyte muscle', and the like, so that the multi-cluster audience characteristic stream can be used as data input for parallel calculation, and the adaptability and the accuracy of the prediction system to different crowds are improved. Finally, the multi-cluster audience feature flow provides data support for subsequent adverse reaction prediction, so that the model can be subjected to customized analysis based on different crowd features, and the scientificity and reliability of prediction are improved.

In one possible implementation manner, the audience characteristic prediction is performed on the target cosmetics according to the cosmetic attribute information to obtain a multi-cluster audience characteristic stream, and step S200 further includes step S210, performing audience prediction on the target cosmetics according to the cosmetic attribute information, and determining a target audience group. Specifically, audience prediction determines an audience population of a target cosmetic by analyzing attribute information of the cosmetic. Firstly, the system predicts potential user groups of the cosmetics according to the types, functional characteristics, applicable skin types and other attributes of the products, for example, the oil control foundation liquid is suitable for young people with oily skin, and the anti-aging essence is suitable for women over 40 years old. Then, the system collects the characteristic information of age, sex, skin quality and the like of the target audience, constructs a target audience characteristic library, and further subdivides the population through age, sex and skin quality classification, such as oily skin of 18-25 years old females, sensitive muscles of 30-40 years old males and the like. Finally, the system integrates the classification result into a multi-cluster audience characteristic stream, and provides accurate data support for subsequent personalized adverse reaction prediction, so that the accuracy and pertinence of prediction are improved.

Step S220, collecting age information, sex information and skin information of the target audience group to obtain a target audience feature library. Specifically, a target audience feature library is constructed by collecting age, gender and skin information of the target audience. First, the system predicts a target audience group according to attribute information (such as product category, function, applicable skin, etc.) of cosmetics, and collects age, gender, and skin data of the audience through questionnaire, e-commerce platform analysis, and social media data. For age information, the system classifies the audience into different age groups (e.g., 18-25 years, 26-35 years, etc.), and further refines the audience classification based on gender and skin (e.g., oiliness, dryness, mixedness, sensitive muscles). By integrating the data, the system establishes a database containing audience characteristics, ensuring that subsequent adverse reaction predictions and personalized recommendations can be analyzed based on accurate user data. The process provides important support for safety evaluation of cosmetics, and improves accuracy of product matching and adverse reaction prediction.

And step S230, age and sex classification is carried out according to the target audience feature library, and a multi-cluster audience initial classification result is obtained. Specifically, based on the target audience feature library, the system first performs age-gender classification to determine a target audience population. The system predicts audience by attribute information of the target cosmetics (such as product category, efficacy and applicable skin quality). In terms of age dimension, the system divides the audience into a plurality of age groups, such as 18-25 years old young people, which usually need oil control, moisturizing and basic skin care products, such as oil control powder base solution, moisturizing spray and the like, 26-35 years old young women pay more attention to whitening and anti-aging effects, and are suitable for using whitening essence containing nicotinamide and vitamin C, 36-45 years old women begin to pay attention to anti-aging and tightening skin care products, such as anti-aging essence, tightening cream and the like, and the population over 46 years old usually has loose and dry skin and is biased to use moisturizing and anti-aging repair products. Next, the system refines the classification further based on the gender information. For females, the requirements are quite various, and the requirements from basic skin care to make-up products, anti-aging essences and the like are quite high, while males are usually biased towards basic care, such as oil control, cleaning, sun protection and the like. For example, men typically prefer simple facial cleansing, moisturizing creams, and women choose products that contain a fragrance or more make-up effect. By way of the above, the system divides the target audience into a plurality of groups, such as '18-25 years old, female, oily skin' may tend to choose a refreshing type moisturizing and oil control product, while '36-45 years old, male, sensitive muscle' is more favored as a mild restorative skin care product. And finally, classifying the subdivided audience groups into multi-cluster audience initial classification results by the system, wherein each group has similar characteristics and requirements, so that the follow-up personalized recommendation and adverse reaction prediction are convenient. Classification not only helps determine the potential market for cosmetics, but also provides accurate data support for subsequent accurate predictions and safety assessments.

And step S240, performing skin classification according to the initial classification result of the multi-cluster audience to obtain the multi-cluster audience characteristic stream. Specifically, the system subdivides the target audience according to skin information (such as oiliness, dryness, mixedness and sensitivity) of each audience, so that the most suitable skin care advice can be obtained for groups with different skin types. For example, oily skin audiences are often faced with problems of oily feel, enlarged pores, and acne, and therefore they tend to choose oil-control, refreshing skin care products such as oil-control lotions, light and thin foundations, refreshing creams, and the like. The system groups these populations into clusters, such as "18-25 year old women, oily skin", whose need has focused on oil control and moisturization. For dry skin audiences, the skin lacks moisture and is easy to feel tight and dry, and generally moist and moist products such as creams and essences containing hyaluronic acid, ceramide and other components are required. The system classifies users into a group of '26-35 year old females, dry skin', and preferentially selects high-moisture-retention products when recommending products. For mixed skin populations, the T region is generally greasy and the U region is dry, and such populations prefer products that balance grease and moisture, such as light moisturizing emulsions, oil control essences, and the like. For example, the "36-45 year old male, mixed skin" population recommends the use of balanced products to address skin problems in different areas. Finally, the population of sensitive skin responds strongly to external stimuli, which is liable to cause reddening and allergy, so that they need mild, non-irritating skin care products, such as restorative emulsions, creams with low sensitive ingredients, and the like. The users are classified into 'men over 45 years old, sensitive skin' clusters, and products with mild and strong repair and repairability are recommended. Through skin classification, the system subdivides the audience into different clusters, provides personalized product recommendation for each group, generates multi-cluster audience characteristic streams, provides accurate basis for subsequent adverse reaction prediction, ensures that the skin care requirement of each group is met, and improves the accuracy of cosmetic adverse reaction prediction.

And step S300, performing registration optimization on the cosmetic composition database according to the cosmetic attribute information and the cosmetic composition information, and determining a registration cosmetic composition database. Specifically, the cosmetic composition database registration optimization optimizes and registers the existing cosmetic composition database by analyzing the attribute information and the composition information of the target cosmetics, thereby ensuring the accuracy of the composition data. Firstly, the system screens the component database according to the attribute information of cosmetics, such as product types (such as sun cream, anti-aging essence, face cream and the like), effects (such as whitening, anti-aging, moisturizing and the like) and skin suitable (such as dry skin, oily skin, sensitive skin and the like). For example, for a moisturizing cream, the system first screens out ingredients with moisturizing function, such as hyaluronic acid, glycerin, ceramide, and the like. Then, the system performs twin degree evaluation on each component and the target cosmetic component based on the cosmetic component information, i.e., calculates the similarity of their structure, chemical properties, and efficacy. For example, hyaluronic acid and glycerol generally have a high similarity in moisturizing effect, and thus their registration coefficients are high, and are suitable for use in moisturizing products. Then, the system screens out components with registration coefficients greater than the threshold by setting a component registration threshold. For example, if the target cosmetic is anti-aging concentrate, the system may set a registration factor of 0.8 and only ingredients with registration factors greater than 0.8, such as retinol and peptides, will be selected for the final formulation. Finally, the system generates a registered cosmetic ingredient library which contains all the optimized and screened ingredients, so that the system has higher matching degree and meets the functional requirement of the product. For example, for an anti-aging essence, the registration component library may include hyaluronic acid, retinol, and peptide components that have significant effects on anti-aging and skin repair. Registering the cosmetic ingredient library will serve as an important reference for the prediction of subsequent adverse reactions, personalized recommendation and product development, and provide data support for ensuring the safety and effectiveness of cosmetics.

In a possible implementation manner, the registration and optimization is performed on the cosmetic composition database according to the cosmetic attribute information and the cosmetic composition information, and a registration cosmetic composition database is determined, and step S300 further includes step S310 of performing association screening on the cosmetic composition database according to the cosmetic attribute information to obtain a cosmetic composition database with the same attribute. Specifically, the cosmetic composition database is subjected to association screening according to the attribute information of the target cosmetics, and a composition set matched with the functions, the types and the applicable skin types of the target cosmetics is screened. For example, if the target cosmetic is a moisturizing cream, the system will screen the ingredients in the ingredient database for moisturizing effects, such as hyaluronic acid, glycerin, ceramide, etc., which generally have good water-locking ability, and are suitable for dry and sensitive skin, according to the moisturizing function. If the target cosmetic is anti-aging essence, the system will screen out ingredients with anti-aging efficacy, such as retinol, peptides, etc., which help to stimulate skin regeneration, reduce fine lines and wrinkles. In addition, if the target cosmetic is a sunscreen cream, the components with ultraviolet protection effect, such as titanium dioxide, zinc oxide, avobenzone and the like, can be screened out, can effectively block ultraviolet rays, and reduce the risks of skin aging and sunburn. In the screening process, the system can further refine and screen according to the applicable skin types of the target product, so as to ensure that the components meet the requirements of different skin types. For example, oily skin is suitable for use with oil control ingredients such as salicylic acid, tea tree oil, while dry skin requires more moisturizing ingredients such as squalene, glycerin. Finally, the system screens out all components meeting the requirements of the target cosmetic attributes to form a common-attribute cosmetic component library which contains all components suitable for the target product. The components provide a basis for the follow-up formula optimization and adverse reaction prediction, and ensure the function and safety of the target cosmetics.

And step S320, carrying out twin degree evaluation on each co-attribute cosmetic ingredient data in the co-attribute cosmetic ingredient library according to the cosmetic ingredient information to obtain a plurality of ingredient registration coefficients. Specifically, twin degree evaluation is performed on each component in the co-attribute cosmetic component library using the component information of the target cosmetic to calculate the registration coefficient of each component. To the comprehensive assessment of the chemical structure, functionality and skin reactivity of each component. Taking a moisturizing cream as an example, hyaluronic acid is taken as a common moisturizing component, and is often evaluated to be highly matched because of its simple structure and effective moisturizing, a high registration coefficient, such as 0.9, may be obtained, while glycerin, which is also a moisturizing component, may be slightly weaker in some efficient formulations, although its moisturizing effect is better, and the registration coefficient may be 0.85. In addition, nicotinamide has good moisturizing effect as a component having whitening, anti-aging and anti-inflammatory functions, but has a main function of not moisturizing compared to hyaluronic acid, so that it is possible to obtain a low registration coefficient, e.g., 0.7. And screening out the components which best meet the requirements of the target product according to the registration coefficients of the components, and preferentially selecting the components with high registration coefficients into a final formula. For example, in this type of moisturizing cream, hyaluronic acid and glycerin may be selected, while niacinamide may be excluded. In this way, the system can ensure the high matching of the selected components in functions and effects, and simultaneously provide scientific basis for the subsequent formula optimization and adverse reaction prediction, and ensure the efficacy and safety of the product.

Step S330, optimizing and selecting the plurality of component registration coefficients according to the component registration threshold value to obtain component registration optimizing distribution which is larger than or equal to the component registration threshold value. Specifically, the registration coefficients of each component are first screened according to a set component registration threshold value, ensuring that only those components that highly match the target cosmetic are selected. For example, in developing an anti-aging essence, the functional requirements of the target product include anti-aging and skin repair, so the system sets a registration threshold of 0.8. It is assumed that after twin evaluation, the registration coefficient of hyaluronic acid was 0.9, the registration coefficient of retinol was 0.85, the registration coefficient of glycerin was 0.75, and the registration coefficient of perfume ingredient was 0.3. The system will add the registration coefficients of hyaluronic acid and retinol to the ingredient registration optimizing profile because their registration coefficients are greater than the set threshold value of 0.8, while glycerin and fragrance ingredients will be excluded because their registration coefficients are below the threshold value. Finally, the resulting component registration optimizing profile contains hyaluronic acid and retinol, with registration coefficients of 0.9 and 0.85, respectively, indicating that they are highly matched to the functional requirements of the target anti-aging essence, and can provide the desired moisturizing and anti-aging effects. In this way, the system can effectively screen out the components which best meet the requirements of the target cosmetics and provide powerful support for optimizing the product formulation.

And S340, screening the cosmetic ingredient library with the same attribute according to the ingredient registration optimizing distribution to generate the registration cosmetic ingredient library. Specifically, the registration coefficient of each component is screened according to the component registration threshold value, so that the selected component can be matched with the functional requirement of the target cosmetic highly. Specifically, when developing an anti-aging essence, the system sets a registration threshold of 0.8 according to the requirements of the target product. For example, after prior twin evaluation, the system gave a registration factor of 0.9 for hyaluronic acid, 0.85 for retinol, 0.75 for glycerin, and 0.3 for the fragrance component. Because the registration coefficients of the hyaluronic acid and the retinol are all larger than 0.8, the hyaluronic acid and the retinol meet the requirements of the moisturizing and anti-aging functions of the target products, so that the hyaluronic acid and the retinol can be reserved, and glycerin and perfume components with registration coefficients lower than 0.8 can be eliminated. Finally, the system adds the registration coefficients of hyaluronic acid and retinol to the component registration optimization distribution, forming an optimized library containing highly matched components. Through the above screening process, the system ensures that the final registered cosmetic ingredient library contains only those ingredients highly compatible with the target product, such as hyaluronic acid and retinol, which are effective in providing anti-aging and moisturizing effects, as well as ensuring the efficacy and safety of the product.

And S400, introducing a plurality of adverse reaction prediction learners and an adverse reaction prediction loss analyzer to perform adverse reaction prediction loss optimizing learning on the registered cosmetic ingredient library, and constructing a cosmetic adverse reaction prediction double channel, wherein the cosmetic adverse reaction prediction double channel comprises an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel. Specifically, a plurality of adverse reaction prediction learners and adverse reaction prediction loss analyzers are introduced to perform adverse reaction prediction loss optimizing learning on a component library of the target cosmetics, so that a cosmetic adverse reaction prediction double-channel is built. The dual channels include 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 components such as retinol and perfume components through an immediate adverse reaction prediction channel to predict whether the above components will cause immediate reactions such as skin pricking, allergy, etc. in a short time. If the system finds, from historical data, that the fragrance ingredient has a higher probability of allergic reactions, an immediate warning is given and it may be advisable to reduce the proportion of fragrance ingredient. On the other hand, the system evaluates the effect of the above ingredients in long-term use, for example, retinol may cause skin dryness or irritation in long-term use, through a delayed adverse reaction prediction channel, and thus, the system predicts in the delayed channel, reminds the user to monitor skin condition or adjusts the frequency of use. Through the cooperative work of the two channels, the system can predict the safety of the cosmetics in the use process in real time and for a long time, provide guidance for the optimization and adjustment of the formula, and ensure that the final product is effective and safe in the market.

In one possible implementation manner, a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers are introduced to perform adverse reaction prediction loss optimizing learning on the registration cosmetic component library, and a cosmetic adverse reaction prediction dual-channel is built, wherein the cosmetic adverse reaction prediction dual-channel comprises an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, and step S400 further comprises step S410, and adverse reaction record retrieval is performed according to the registration cosmetic component library to obtain an immediate adverse reaction record library and a delayed adverse reaction record library. Specifically, adverse reaction record search is first performed based on a registered cosmetic ingredient library, thereby obtaining immediate and delayed adverse reaction records for each ingredient. Firstly, the system establishes an immediate adverse reaction record base according to historical data, consumer feedback and clinical test data of the components, and records possible reactions of the components in a short period (such as minutes to hours). For example, fragrance ingredients are often reported to elicit immediate reactions such as skin allergies, redness, stinging, etc., all of which data will be collected and stored in an immediate adverse reaction record repository. The system then creates a library of delayed adverse reaction records based on the long term usage data and product trial records for recording the reactions that may be initiated by those cosmetic ingredients during long term usage, such as skin dryness, pigmentation, or desquamation. For example, retinol is effective against aging, but long-term use may lead to skin dryness or peeling, and the above reaction will be recorded in a delayed adverse reaction recording library. Through the comprehensive retrieval of the immediate adverse reaction and the delayed adverse reaction of the components, the system can better understand the risks possibly brought by each component in different using stages, thereby providing scientific basis for optimizing and evaluating the safety of the product formula.

And step S420, collecting audience characteristics according to the immediate adverse reaction record library, and obtaining an audience characteristic first record library. Specifically, firstly, relevant audience characteristics are extracted from an immediate adverse reaction record library, and a first record library of the audience characteristics is established by collecting information such as age, gender, skin quality, allergy history and the like of a user. For example, assuming that a skin care product containing a fragrance component produces an allergic reaction in an oily skin population of a female aged 20-30 years, the system will extract the characteristic information (age, sex, skin, type of reaction) of the population and store in a first record of audience characteristics. Also, if some users develop immediate reactions such as dry skin or stinging when using anti-aging essence containing retinol, the system will record the immediate reactions and correlate the data with corresponding audience characteristics (e.g., dry skin of 30-40 year old women, allergy history, etc.). For each audience group, the system also collects information of skin types (such as oiliness, dryness, sensitivity and the like) and whether allergy histories exist, so that subsequent predictions are more personalized. Through the characteristics, the system can establish an accurate first record library of audience characteristics according to the type of immediate adverse reaction and the specific information of the audience, and provides powerful data support for subsequent cosmetic formula optimization and adverse reaction prediction.

And step S430, collecting audience characteristics according to the delayed adverse reaction record library, and obtaining a second record library of the audience characteristics. Specifically, audience characteristics related to long-term use are extracted from a delayed adverse reaction record base, and characteristic data are stored in an audience characteristic second record base. For example, assuming that a retinoid-containing anti-aging concentrate causes dryness and peeling after prolonged use, the system would record data on delayed responses and correlate them with age, sex, skin characteristics, etc. of the particular audience. Female users of 30-40 year old dry skin may develop significant skin dryness or peeling when using retinol, and thus the data is recorded as a specific audience characteristic. Also, for 20-30 year old oily skin men using the same product, the system may not record significant delayed adverse effects, as oily skin users typically have a higher tolerance to retinol. The system may also collect other characteristics of different audience populations, such as whether there is a history of allergies, frequency of use, etc., particularly for long term use of ingredients that may cause pigmentation, allergic reactions, such as products containing certain preservatives or fragrances. Through the data, the system can generate a complete audience characteristic second record library, so that a more accurate basis is provided for subsequent risk prediction, and the safety of the product in long-term use is ensured.

Step S440, according to the adverse reaction prediction learners and the adverse reaction prediction loss analyzer, performing instant adverse reaction prediction loss optimizing learning on the registration cosmetic component library, the audience characteristic first record library and the instant adverse reaction record library, and generating the instant adverse reaction prediction channel. Specifically, the instantaneous adverse reaction prediction loss optimizing learning is performed on the registered cosmetic ingredient library, the audience characteristic first record library and the instantaneous adverse reaction record library through a plurality of adverse reaction prediction learners (machine learning models) and an adverse reaction prediction loss analyzer, so that an instantaneous adverse reaction prediction channel is generated. For example, the system uses a decision tree to analyze whether certain components (e.g., fragrances) may elicit allergic reactions in a particular audience (e.g., women aged 30-40), and through training of a machine learning model, the system can identify which components may elicit allergies in users with oily skin and less reactions in users with dry skin. The support vector machine helps the system handle complex relationships between different components and user features, and in particular classifies immediate responses under the combined action of multiple features (such as skin, allergy history, age, etc.). In the model training process, the loss analyzer optimizes the learner by calculating the error between the predicted result and the actual reaction. For example, if a fragrance ingredient elicits an unpredictable allergic response in certain user populations, the loss resolver will adjust the model, reduce such errors, and readjust the usage proportions of the ingredient. Through the process, a plurality of learners finally generate an immediate adverse reaction prediction channel, and immediate adverse reactions such as allergy, stinging, red swelling and the like after the cosmetics are used can be predicted in real time.

And S450, performing delayed adverse reaction prediction loss optimizing learning on the registration cosmetic ingredient library, the audience characteristic second record library and the delayed adverse reaction record library according to the plurality of adverse reaction prediction learners and the adverse reaction prediction loss analyzer, and generating the delayed adverse reaction prediction channel. Specifically, the delayed adverse reaction prediction loss optimizing learning is performed on the registered cosmetic ingredient library, the audience characteristic second record library and the delayed adverse reaction record library through a plurality of adverse reaction prediction learners and the adverse reaction prediction loss analyzer so as to generate a delayed adverse reaction prediction channel. For example, when the system analyzes such components as retinol, multiple machine learning models (e.g., decision trees, neural networks, random forests) may be used to learn the long term effects of retinol on users of different ages and skin conditions. Given the dry skin user over 40 years old, the system may find that the retinol component may cause delayed reactions such as skin dryness, peeling, etc. after prolonged use. After the model inputs the data, the system can identify which age groups, gender and skin are most likely to have side effects by further analyzing the data through a support vector machine. In this process, the loss resolver optimizes the prediction results of the models by calculating the error of each model. For example, if the system predicts that the delay adverse reaction possibly caused by a certain component does not match with the actual data, the loss analyzer will adjust according to the error, so as to reduce the prediction deviation and improve the accuracy. By the mode, the system continuously optimizes the prediction model, generates a final delayed adverse reaction prediction channel, and can early warn side effects which can occur after long-term use, such as skin dryness, pigmentation, long-term allergy and the like, so that data support is provided for safety optimization of a product formula.

Step S460, connecting 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 firstly generates an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, and the adverse reactions of the cosmetic ingredients in short-term and long-term use are respectively and independently analyzed. The immediate adverse reaction prediction channel focuses on short-term reactions such as immediate adverse reactions such as allergies, stinging or redness and swelling that the perfume ingredients may cause in users of oily skin. Through multiple machine learning models (e.g., decision trees and support vector machines), the system learns and predicts these short-term responses, issues alerts in real-time, and provides recipe adjustment suggestions. On the other hand, the delayed adverse reaction prediction channel aims at adverse reactions after long-term use, such as skin dryness, desquamation or pigmentation possibly caused by retinol components during 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 advice, such as reducing frequency of use or changing components. In order to improve the efficiency and the prediction precision of the system, the two channels are connected in a parallel node mode, so that the instant and delayed reactions can be processed simultaneously and predicted independently, and finally a cosmetic adverse reaction prediction double channel is generated. The dual-channel system not only can timely identify short-term adverse reactions such as anaphylactic reactions caused by perfume, but also can predict long-term risks such as dryness or desquamation of dry skin caused by retinol, thereby comprehensively evaluating the safety of cosmetics and providing accurate suggestions for optimizing products.

In one possible implementation manner, according to the multiple adverse reaction prediction learners and the adverse reaction prediction loss analyzer, performing instantaneous adverse reaction prediction loss optimizing learning on the registered cosmetic component library, the audience feature first record library and the instantaneous adverse reaction record library, and generating the instantaneous adverse reaction prediction channel, step S440 further includes step S441, using the registered cosmetic component library and the audience feature first record library as input information and the instantaneous adverse reaction record library as output information, respectively performing supervision training on the multiple adverse reaction prediction learners, and calculating multiple instantaneous adverse reaction prediction loss coefficients according to the adverse reaction prediction loss analyzer when training is performed for a predetermined training number. Specifically, the system uses the registered cosmetic ingredient library and the audience characteristic first record library as input data, and uses the immediate adverse reaction record library as output information to perform multiple training to generate an immediate adverse reaction prediction model. First, a plurality of adverse reaction prediction learners (e.g., decision trees, support vector machines, neural networks, etc.) are used to train models to predict which components are likely to induce allergic, tingling, redness, etc. reactions in a short period of time by learning the relationship between the cosmetic components and the audience characteristics. For example, the fragrance component may cause allergy to oily skin users with less impact on dry skin users. In each training process, the system calculates the instantaneous adverse prediction loss coefficient according to the adverse reaction prediction loss analyzer, and evaluates the error between the prediction result and the actual reaction. Assuming that the model predicts that a component (e.g., preservative) will trigger an allergic reaction in 20% of the users, but the actual data indicates that only 15% of the users react, the loss resolver will calculate an error and adjust the model parameters. If the loss coefficient is lower than a preset instant adverse reaction prediction loss threshold value, the system considers that the model has reached the required accuracy and generates a plurality of instant adverse reaction prediction models. The models are connected together through an ensemble learning method (such as a weighted average or voting mechanism) to form an instantaneous adverse reaction prediction channel. For example, for a certain cosmetic product, the system may predict immediate allergic reactions of the fragrance ingredient in an oily skin user, while predicting the potentially stinging sensation of the preservative ingredient in the sensitive skin, ultimately providing safety assessment and formulation adjustment advice to the developer.

In step S442, if the plurality of immediate adverse prediction loss coefficients is smaller than the immediate adverse prediction loss threshold, a plurality of immediate adverse reaction prediction models are generated. Specifically, the system uses a registered cosmetic ingredient library and an audience characteristic first record library as input data, an immediate adverse reaction record library as output data, and performs supervision training through a plurality of adverse reaction prediction learners (such as decision trees, support vector machines, neural networks and the like). With each training, the system uses the data in the training set (including cosmetic ingredients and user characteristics) to predict the immediate response of cosmetic ingredients to the possible occurrence of different populations (e.g., oily skin, dry skin, or sensitive skin users), such as allergies, redness, or stings. By training, the model will gradually optimize, identifying which components are more likely to elicit these responses in a particular audience population. For example, fragrance ingredients may cause stinging or allergic reactions in dry skin users, whereas such reactions may not occur in oily skin users. The system calculates the loss coefficient of each training through the adverse reaction prediction loss analyzer, and measures the error between the prediction result and the actual reaction. If the instantaneous adverse prediction loss coefficients are smaller than the set instantaneous adverse prediction loss threshold, the system considers the model to have reached the expected accuracy, and a plurality of instantaneous adverse reaction prediction models are generated. For example, the system may generate a decision tree model that identifies the immediate response of the fragrance component to different ages and skin conditions and adjusts its predictions based on past training data. And then, the system connects the models through an ensemble learning method (such as a weighted average or voting mechanism) to form a final instant adverse reaction prediction channel, and the channel can predict the instant adverse reaction possibly caused by the cosmetic ingredients in the use process in real time and provide accurate early warning for the formula optimization and the safety evaluation of the product.

Step S443, connecting the plurality of instant adverse reaction prediction models to obtain the instant adverse reaction prediction channel. Specifically, in the cosmetic adverse reaction prediction system, when a plurality of immediate adverse reaction prediction models are generated after training, the system connects the models through an integrated learning method to form an immediate adverse reaction prediction channel. For example, assume that there are three models, a decision tree, a Support Vector Machine (SVM), and a neural network, each model learning the relationship between the cosmetic composition and the audience based on different characteristics. The decision tree model may find that the perfume ingredients are more allergic to oily skin users, while the SVM model may predict that the perfume is more irritating to sensitive skin, and the neural network provides comprehensive predictions based on more dimensional data (e.g., age, allergy history, etc.). The system integrates the prediction results of the model by means of weighted averaging or voting mechanism. For example, if both the decision tree and neural network models predict that a particular fragrance component will cause an allergic reaction in a dry skin user, and the SVM model does not predict such a reaction, the system will make a final decision based on the predictions of the majority model. In addition, the performance of each model in the training process determines the weight of each model in the integration, and the model performs better and can obtain higher weight. Finally, by integrating the model, the immediate adverse reaction prediction channel can comprehensively evaluate the immediate reactions, such as redness, tingling, allergy, etc., which can be caused by cosmetics in user groups with different skin types, ages and allergy histories. For example, when a particular fragrance-containing cream is predicted to cause redness and swelling in a female user of sensitive skin, the system may provide feedback suggesting reduced fragrance use or adjusted ingredient proportions to optimize the product formulation and improve its safety.

In one possible implementation manner, introducing a plurality of adverse reaction prediction learners and an adverse reaction prediction loss analyzer to perform adverse reaction prediction loss optimizing learning on the registered cosmetic ingredient library, and building a cosmetic adverse reaction prediction dual channel, wherein the cosmetic adverse reaction prediction dual channel comprises an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel, step S400 further comprises step S470, and the adverse reaction prediction loss analyzer comprises an adverse reaction prediction loss analysis function, and the adverse reaction prediction loss analysis function is: Wherein LOSS represents an adverse prediction LOSS coefficient, M represents preset training times, M represents M-th training, M and M are positive integers, M is not less than 1 and not more than M, SUO _m represents the number of adverse reaction prediction samples in M-th training, and SUX _m represents the number of adverse reaction prediction correct samples in M-th training. Specifically, in the cosmetic adverse reaction prediction system, the adverse reaction prediction loss resolver evaluates accuracy of model prediction by calculating a loss coefficient. In each training process, the system calculates the loss coefficient of each training round according to an adverse reaction prediction loss analysis function, wherein the form of the function is as follows: Where SUO _m represents the number of adverse reaction predicted samples in the mth training round, SUX _m represents the number of correctly predicted samples, and M is the number of training rounds. For example, in a certain round of training, if the system processed 100 samples and successfully predicted 80 correct responses, the loss factor was 0.2, indicating that the prediction accuracy of the model was to be improved. After each training round, the system evaluates the model performance according to the calculated loss coefficient, if the loss coefficient is smaller than a preset loss threshold value, the model prediction is accurate enough, and a final instant adverse reaction prediction model is generated by a plurality of models. For example, assuming that the system is trained to find perfume ingredients that will cause allergic reactions in users of certain ages, after calculation by the loss resolver, if the loss coefficient is still above the threshold, the system will adjust model parameters, such as reducing perfume usage or performing different formulation optimizations for sensitive people, until the loss coefficient meets the requirements, and the resulting predictive model will accurately reflect the immediate response of perfume to user populations of different skin, age and allergy history. The process ensures that potential adverse reactions in cosmetics can be identified as accurately as possible by continuously optimizing loss coefficients, thereby providing scientific basis for the safety of products.

And S500, inputting the cosmetic ingredient information and the multi-cluster audience characteristic stream into the cosmetic adverse reaction prediction double channel to generate a cosmetic adverse reaction prediction report. Specifically, in the system for predicting the adverse reaction of cosmetics, the system uses the information of the components of cosmetics and the characteristic streams of multiple clusters of audience as input data, and transmits the input data to the dual channels for predicting the adverse reaction of cosmetics for processing. First, the system predicts both short term (immediate adverse effects) and long term (delayed adverse effects) based on cosmetic composition information, such as hyaluronic acid, retinol, preservatives, fragrances, etc., in combination with multiple clusters of audience characteristics streams, including skin quality, age, gender, allergy history, etc., of the user. For example, the system will analyze the immediate response of the fragrance component to an oily skin user, predicting the likely allergic response to be elicited, and at the same time will evaluate the effect of the retinol component after prolonged use, especially the risk of skin dryness or peeling in dry skin users. The immediate adverse reaction prediction channel can treat red swelling, allergy or stinging and other reactions possibly caused by the perfume components, and can make different predictions for different skin types and age groups. The delayed adverse reaction prediction channel predicts problems such as skin dryness, pigmentation, etc. which may be caused by long-term use of retinol by analyzing the long-term use effect of cosmetic ingredients. Finally, the system combines the predictions of the two channels to generate an exhaustive cosmetic adverse reaction prediction report that lists the short-term and long-term reactions of each component in different user populations. For example, fragrance ingredients may cause allergies in female users of sensitive skin, while retinol may cause dryness or desquamation in dry skin users over 40 years old. Through such a two-channel process, the developer can adjust the product formulation, e.g., reduce the use concentration of fragrance, or adjust the retinol content, based on the report, thereby optimizing the safety of the product and ensuring its applicability in the market.

In one possible implementation manner, the cosmetic composition 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, and step S500 further includes step S510 of inputting the cosmetic composition 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. Specifically, in the system for predicting the adverse reaction of cosmetics, the component information of cosmetics and the characteristic flow of multiple clusters of audience are taken as input and enter an immediate adverse reaction prediction channel for processing. First, the system analyzes cosmetic composition information, such as perfume, retinol, preservative, etc., in combination with multiple clusters of audience characteristics streams, including information on the user's age, gender, skin type (e.g., oily skin, dry skin, sensitive skin), and allergy history. The system evaluates the immediate response of these components in different user populations through multiple machine learning models (e.g., decision trees, SVMs, neural networks, etc.). For example, fragrance ingredients may elicit allergic reactions in oily skin populations, but hardly react in dry skin users. The system can accurately predict the immediate adverse reaction of different components to each group, such as red swelling, tingling, allergy and the like by training a model. If the decision tree model predicts 80% of the allergy probability of the perfume ingredient in the oily skin population and the neural network model predicts 60% of the red and swollen response probability of the ingredient in the sensitive skin user, the system will generate immediate adverse reaction prediction results for each audience population based on the predictions of these different models. These results will help the research and development team identify immediate allergic reactions that cosmetic ingredients may elicit in different user groups, providing scientific basis for formulation adjustment of the product. Finally, the multi-cluster audience-immediate poor prediction results output by the system will detail the risk of immediate reactions that may be initiated by specific cosmetic ingredients in each audience population, thereby providing data support for product optimization and safety assessment.

And step S520, inputting the cosmetic ingredient information and the multi-cluster audience characteristic flow into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse prediction result. In particular, cosmetic composition information and multi-cluster audience feature streams are input into a delayed adverse reaction prediction channel, the primary task of which is to evaluate the potential impact of cosmetic compositions on different populations after prolonged use. The system first analyzes information about cosmetic ingredients, such as retinol, hyaluronic acid, preservatives, etc., which may have an effect on the skin after several weeks or months. For example, retinol may cause desquamation or dryness to dry skin users, but has less impact on oily skin populations. The multi-cluster audience feature stream, in turn, provides detailed information about the individual user population, including age, gender, skin quality (e.g., oily, dry, sensitive skin), allergy history, and the like. Based on this information, the system predicts the potential response of the different components in long term use through multiple machine learning models (e.g., decision trees, SVMs, neural networks, etc.). For example, for a dry skin population over 40 years old, the system might predict a 60% probability of retinol ingredients initiating skin dryness or desquamation, whereas for a younger oily skin population, this probability might be only 20%. The system will calculate the risk of delayed reactions, such as pigmentation, dry skin or allergies, for each cosmetic ingredient in each population based on the characteristics of that population. Finally, the system outputs multi-cluster audience-delayed adverse reaction prediction results, provides personalized delayed adverse reaction prediction for each group, and helps research and development teams evaluate the safety of the components and optimize the product formulation. For example, if the system detects that retinol may trigger a peeling reaction in a dry skin population, the research and development team may choose to reduce the concentration of that ingredient or such users develop alternative products that do not contain retinol.

And step S530, carrying out data fusion according to the multi-cluster audience-instant adverse prediction result and the multi-cluster audience-delayed adverse prediction result, and outputting the cosmetic adverse reaction prediction report. Specifically, the predicted result of the immediate adverse reaction and the predicted report of the delayed adverse reaction are generated by data fusion, and the adverse reaction possibly caused by each component in a short term (immediate reaction) and a long term (delayed reaction) is respectively evaluated according to the characteristics (such as skin quality, age, sex, allergy history and the like) of different audience groups and the information of the components of the cosmetics. For example, fragrance ingredients may trigger a higher immediate allergic reaction in oily skin users, while retinol ingredients may trigger dryness and desquamation after prolonged use in dry skin populations. The immediate adverse reaction prediction channel and the delayed adverse reaction prediction channel predict the reactions respectively to generate respective results. The system then fuses the two types of predictions through a weighted average or voting mechanism. For example, the immediate response of a fragrance ingredient in an oily skin user is predicted to be 80%, while the delayed response in dry skin is predicted to be 20%. The system gives different weights according to the group characteristics through weighted calculation to obtain comprehensive evaluation. If the fragrance ingredient has a higher immediate risk of reacting in the sensitive skin population and retinol has a higher delayed risk of reacting in the dry skin population, the report alerts the two types of ingredients, respectively, prompting the research team to reduce the fragrance ingredient or adjust the concentration of retinol. In addition, the system comprehensively considers the characteristics of each population through a decision tree method, and the finally generated report details the instant and delayed reaction risk of each population and provides suggestions for optimizing the formulation, such as suggestions for sensitive skin users to reduce the use of fragrances, or alternatives without retinol for dry skin users.

According to the embodiment of the application, the attribute and component information of the target cosmetics are acquired, and then the audience characteristics of the target cosmetics are predicted according to the information, so that a multi-cluster audience characteristic stream is formed. And performing registration optimization on the cosmetic composition database by using the cosmetic attributes and the composition information, and determining a registration composition database. And learning by adopting a plurality of adverse reaction prediction learners and a loss analyzer, and constructing a dual-channel for instant and delayed adverse reaction prediction. The components of the cosmetics and the audience characteristics are input into the double channels to generate a prediction report of the adverse reaction of the cosmetics, so that the personalized and accurate prediction of the adverse reaction of the cosmetics is realized, and the technical effect of improving the prediction accuracy is achieved.

Hereinabove, a cosmetic composition database-driven cosmetic adverse reaction prediction method according to an embodiment of the present invention is described in detail with reference to fig. 1. Next, a cosmetic adverse reaction prediction system driven by a cosmetic composition database according to an embodiment of the present invention will be described with reference to fig. 2.

The cosmetic adverse reaction prediction system driven by the cosmetic composition database provided by the embodiment of the invention is used for solving the technical problem that personalized and accurate prediction cannot be provided in the existing cosmetic adverse reaction prediction, and achieves the technical effect of improving the prediction accuracy by realizing personalized and accurate prediction of the cosmetic adverse reaction. The cosmetic composition database-driven cosmetic adverse reaction prediction system comprises a prediction instruction acquisition module 10, an audience characteristic prediction module 20, a database registration optimizing module 30, a prediction double-channel construction module 40 and an adverse reaction prediction report generation module 50.

The prediction instruction acquisition module is used for 10 obtaining a prediction instruction of the adverse reaction of the cosmetics, wherein the prediction instruction of the adverse reaction of the cosmetics comprises cosmetic attribute information and cosmetic ingredient information corresponding to a target cosmetic.

The audience characteristic prediction module 20 is configured to perform audience characteristic prediction on the target cosmetic according to the cosmetic attribute information, so as to obtain a multi-cluster audience characteristic stream.

The database registration optimizing module 30 is configured to perform registration optimizing on the cosmetic composition database according to the cosmetic attribute information and the cosmetic composition information, and determine a registered cosmetic composition database.

The predicted dual-channel building module 40 is configured to introduce a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers to perform adverse reaction prediction loss optimizing learning on the registered cosmetic ingredient library, and build a cosmetic adverse reaction prediction dual-channel, wherein the cosmetic adverse reaction prediction dual-channel comprises an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel.

The adverse reaction prediction report generation module 50 is configured to input the cosmetic composition information and the multi-cluster audience feature stream into the cosmetic adverse reaction prediction dual channel, and generate a cosmetic adverse reaction prediction report.

Next, a specific configuration of the audience feature prediction module 20 will be described in detail. As described above, the audience characteristic prediction module 20 further includes a target audience group determination unit configured to perform audience prediction on the target cosmetics according to the cosmetic attribute information, determine a target audience group, a target audience characteristic library acquisition unit configured to acquire age information, gender information, and skin information of the target audience group, and obtain a target audience characteristic library, an initial classification result acquisition unit configured to perform age-gender classification according to the target audience characteristic library, and obtain a multi-cluster audience initial classification result, and a skin classification unit configured to perform skin classification according to the multi-cluster audience initial classification result, and obtain the multi-cluster audience characteristic stream.

Next, the specific configuration of the database registration optimizing module 30 will be described in detail. As described above, the database registration optimizing module 30 further includes a cosmetic ingredient library obtaining unit configured to perform association screening on the cosmetic ingredient database according to the cosmetic attribute information to obtain a co-attribute cosmetic ingredient library, a twin degree evaluating unit configured to perform twin degree evaluation on each co-attribute cosmetic ingredient data in the co-attribute cosmetic ingredient library according to the cosmetic ingredient information to obtain a plurality of ingredient registration coefficients, an optimizing selecting unit configured to perform optimizing on the plurality of ingredient registration coefficients according to an ingredient registration threshold to obtain an ingredient registration optimizing distribution equal to or greater than the ingredient registration threshold, and a cosmetic ingredient library screening unit configured to screen the co-attribute cosmetic ingredient library according to the ingredient registration optimizing distribution to generate the cosmetic ingredient.

Next, the specific configuration of the predictive two-channel construction module 40 will be described in detail. As described above, introducing a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers to perform adverse reaction prediction loss optimizing learning on the registered cosmetic ingredient library, setting up a cosmetic adverse reaction prediction dual-channel, wherein the cosmetic adverse reaction prediction dual-channel comprises an instantaneous adverse reaction prediction channel and a delayed adverse reaction prediction channel, the prediction dual-channel setting up module 40 further comprises an adverse reaction record library acquisition unit for performing adverse reaction record retrieval according to the registered cosmetic ingredient library to obtain an instantaneous adverse reaction record library and a delayed adverse reaction record library, an audience characteristic first record library acquisition unit for performing audience characteristic acquisition according to the instantaneous adverse reaction record library to obtain an audience characteristic first record library, an audience characteristic second record library acquisition unit for performing audience characteristic acquisition according to the delayed adverse reaction record library to obtain an audience characteristic second record, an audience characteristic second record library, a predicted channel generating unit for generating the instantaneous adverse reaction prediction channel and the instantaneous adverse reaction prediction request, and the delay adverse reaction prediction unit for generating the instantaneous adverse reaction record and the predicted adverse reaction loss analysis by the instantaneous adverse reaction analysis unit, the delayed adverse reaction prediction channel generation unit is used for carrying out delayed adverse reaction prediction loss optimizing learning on the registration cosmetic ingredient library, the audience characteristic second record library and the delayed adverse reaction record library according to the plurality of adverse reaction prediction learners and the adverse reaction prediction loss analyzer to generate the delayed adverse reaction prediction channel, and the parallel node connection unit is used for connecting the instant adverse reaction prediction channel and the delayed adverse reaction prediction channel as parallel nodes to generate the cosmetic adverse reaction prediction dual channel.

The system comprises a registration cosmetic component library, an audience characteristic first record library, an immediate adverse reaction prediction loss analysis device, an adverse reaction prediction model generation subunit, an immediate adverse reaction prediction channel generation subunit, an adverse reaction prediction model generation subunit and an adverse reaction channel immediate prediction module generation subunit, wherein the registration cosmetic component library, the audience characteristic first record library and the immediate adverse reaction prediction analysis device are subjected to immediate adverse reaction prediction loss optimization learning according to the plurality of adverse reaction prediction learning devices and the adverse reaction prediction loss analysis device, the immediate adverse reaction prediction channel generation subunit is used for generating a plurality of immediate adverse reaction prediction loss thresholds according to the immediate adverse reaction prediction model generation subunit, the loss coefficient calculation subunit is used for taking the registration cosmetic component library and the audience characteristic first record library as input information, the immediate adverse reaction prediction learning devices are respectively subjected to supervision training according to the immediate adverse reaction prediction loss analysis device, and each training preset training time is used for calculating a plurality of immediate adverse reaction prediction loss coefficients according to the adverse reaction prediction loss analysis device, and the adverse reaction prediction model generation subunit is used for acquiring the immediate adverse reaction prediction model of the immediate adverse reaction prediction channel when the plurality of immediate adverse reaction prediction loss coefficients are smaller than the immediate adverse reaction prediction threshold, and the adverse reaction prediction channel generation subunit is used for acquiring the immediate adverse reaction prediction model.

The predicted two-channel building module 40 further includes an adverse reaction predicted loss analyzer component unit, where the adverse reaction predicted loss analyzer component unit is configured to include an adverse reaction predicted loss analysis function, and the adverse reaction predicted loss analysis function is: Wherein LOSS represents an adverse prediction LOSS coefficient, M represents preset training times, M represents M-th training, M and M are positive integers, M is not less than 1 and not more than M, SUO _m represents the number of adverse reaction prediction samples in M-th training, and SUX _m represents the number of adverse reaction prediction correct samples in M-th training.

Next, the specific configuration of the adverse-reaction prediction report generation module 50 will be described in detail. As described above, the cosmetic composition 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 comprises a multi-cluster audience-immediate adverse reaction prediction result acquisition unit for inputting the cosmetic composition information and the multi-cluster audience feature stream into the immediate adverse reaction prediction channel to obtain a multi-cluster audience-immediate adverse reaction prediction result, a multi-cluster audience-delayed adverse reaction prediction result acquisition unit for inputting the cosmetic composition information and the multi-cluster audience feature stream into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse reaction prediction result, and a data fusion unit for fusing the data according to the multi-cluster audience-immediate adverse reaction prediction result and the multi-cluster audience-delayed adverse reaction prediction result.

The cosmetic adverse reaction prediction system driven by the cosmetic composition database provided by the embodiment of the invention can execute the cosmetic adverse reaction prediction method driven by the cosmetic composition database provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

Although the present application makes various references to certain modules in a system according to an embodiment of the present application, any number of different modules may be used and run on a user terminal and/or a server, and each unit and module included are merely divided according to functional logic, but are not limited to the above-described division, so long as the corresponding functions can be implemented, and in addition, specific names of each functional unit are only for convenience of distinguishing from each other, and are not intended to limit the scope of protection of the present application.

The above embodiments do not limit the scope of the present application. It will be apparent to those skilled in the art that various modifications, combinations, and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of the present application.

Claims (8)

1. A cosmetic composition database-driven cosmetic adverse reaction prediction method, characterized in that the method comprises:

obtaining a cosmetic adverse reaction prediction instruction, wherein the cosmetic adverse reaction prediction instruction comprises cosmetic attribute information and cosmetic component information corresponding to a target cosmetic;

performing audience characteristic prediction on the target cosmetics according to the cosmetic attribute information to obtain multi-cluster audience characteristic streams;

Performing registration optimization on a cosmetic composition database according to the cosmetic attribute information and the cosmetic composition information, and determining a registration cosmetic composition database;

Introducing a plurality of adverse reaction prediction learners and an adverse reaction prediction loss analyzer to perform adverse reaction prediction loss optimizing learning on the registration cosmetic ingredient library, and constructing a cosmetic adverse reaction prediction double channel, wherein the cosmetic adverse reaction prediction double channel comprises an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel;

And inputting the cosmetic ingredient information and the multi-cluster audience characteristic stream into the cosmetic adverse reaction prediction double channel to generate a cosmetic adverse reaction prediction report.

2. The method of claim 1, wherein performing audience feature prediction on the target cosmetic based on the cosmetic attribute information to obtain a multi-cluster audience feature stream, comprising:

performing audience prediction on the target cosmetics according to the cosmetic attribute information, and determining a target audience group;

collecting age information, sex information and skin information of the target audience group to obtain a target audience feature library;

age and sex classification is carried out according to the target audience feature library, and a multi-cluster audience initial classification result is obtained;

and performing skin classification according to the initial classification result of the multi-cluster audience to obtain the multi-cluster audience characteristic stream.

3. The method of claim 1, wherein performing registration optimization on a cosmetic composition database based on the cosmetic attribute information and the cosmetic composition information, determining a registered cosmetic composition library, comprises:

Performing association screening on the cosmetic ingredient database according to the cosmetic attribute information to obtain a cosmetic ingredient database with the same attribute;

carrying out twin degree evaluation on each co-attribute cosmetic ingredient data in the co-attribute cosmetic ingredient library according to the cosmetic ingredient information to obtain a plurality of ingredient registration coefficients;

Optimizing and selecting the plurality of component registration coefficients according to a component registration threshold value to obtain component registration optimizing distribution which is larger than or equal to the component registration threshold value;

and screening the cosmetic ingredient library with the same attribute according to the ingredient registration optimizing distribution to generate the registration cosmetic ingredient library.

4. The method of claim 1, wherein introducing a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers to perform adverse reaction prediction loss optimizing learning on the registered cosmetic ingredient library, constructing a cosmetic adverse reaction prediction dual channel, comprising:

according to the registration cosmetic ingredient library, carrying out adverse reaction record retrieval to obtain an immediate adverse reaction record library and a delayed adverse reaction record library;

collecting audience characteristics according to the immediate adverse reaction record library to obtain a first record library of audience characteristics;

collecting audience characteristics according to the delayed adverse reaction record library to obtain a second record library of audience characteristics;

according to the adverse reaction prediction learners and the adverse reaction prediction loss analyzer, carrying out instant adverse reaction prediction loss optimizing learning on the registration cosmetic component library, the audience characteristic first record library and the instant adverse reaction record library to generate the instant adverse reaction prediction channel;

According to the adverse reaction prediction learners and the adverse reaction prediction loss analyzer, carrying out delayed adverse reaction prediction loss optimizing learning on the registration cosmetic component library, the audience characteristic second record library and the delayed adverse reaction record library to generate the delayed adverse reaction prediction channel;

and connecting 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.

5. The method of claim 4, wherein performing immediate adverse reaction prediction loss optimization learning on the registered cosmetic composition library, the audience feature first record library, and the immediate adverse reaction record library according to the plurality of adverse reaction prediction learners and the adverse reaction prediction loss parser, generating the immediate adverse reaction prediction channel comprises:

Taking the registration cosmetic component library and the audience characteristic first record library as input information, taking the immediate adverse reaction record library as output information, respectively performing supervision training on the adverse reaction prediction learners, and calculating a plurality of immediate adverse prediction loss coefficients according to the adverse reaction prediction loss analyzer when training for preset training times;

if the instantaneous adverse prediction loss coefficients are smaller than the instantaneous adverse prediction loss threshold, generating a plurality of instantaneous adverse reaction prediction models;

And connecting the plurality of instant adverse reaction prediction models to obtain the instant adverse reaction prediction channel.

6. The method of claim 1, wherein the adverse reaction predicted loss resolver comprises an adverse reaction predicted loss resolution function that is:

The LOSS represents an adverse prediction LOSS coefficient, M represents preset training times, M represents M-th training, M and M are positive integers, M is not less than 1 and not more than M, SUOm represents the number of adverse reaction prediction samples in M-th training, and SUXm represents the number of adverse reaction prediction correct samples in M-th training.

7. The method of claim 1, wherein inputting the cosmetic composition information and the multi-cluster audience feature stream into the cosmetic adverse reaction prediction dual channel generates a cosmetic adverse reaction prediction report, comprising:

Inputting the cosmetic ingredient information and the multi-cluster audience characteristic stream into the immediate adverse reaction prediction channel to obtain a multi-cluster audience-immediate adverse prediction result;

inputting the cosmetic ingredient information and the multi-cluster audience characteristic stream into the delayed adverse reaction prediction channel to obtain a multi-cluster audience-delayed adverse prediction result;

And carrying out data fusion according to the multi-cluster audience-instant adverse prediction result and the multi-cluster audience-delayed adverse prediction result, and outputting the cosmetic adverse reaction prediction report.

8. A cosmetic composition database-driven cosmetic adverse reaction prediction system for implementing the cosmetic composition database-driven cosmetic adverse reaction prediction method according to any one of claims 1 to 7, the system comprising:

The system comprises a prediction instruction acquisition module, a prediction instruction generation module and a control module, wherein the prediction instruction acquisition module is used for acquiring a prediction instruction of adverse reaction of cosmetics, and the prediction instruction of adverse reaction of cosmetics comprises cosmetic attribute information and cosmetic ingredient information corresponding to a target cosmetic;

The audience characteristic prediction module is used for performing audience characteristic prediction on the target cosmetics according to the cosmetic attribute information to obtain a plurality of clusters of audience characteristic streams;

The database registration optimizing module is used for carrying out registration optimizing on the cosmetic composition database according to the cosmetic attribute information and the cosmetic composition information, and determining a registration cosmetic composition database;

The prediction double-channel construction module is used for introducing a plurality of adverse reaction prediction learners and adverse reaction prediction loss resolvers to perform adverse reaction prediction loss optimizing learning on the registration cosmetic ingredient library and constructing cosmetic adverse reaction prediction double channels, wherein the cosmetic adverse reaction prediction double channels comprise an immediate adverse reaction prediction channel and a delayed adverse reaction prediction channel;

And the adverse reaction prediction report generation module is used for inputting the cosmetic ingredient information and the multi-cluster audience characteristic stream into the cosmetic adverse reaction prediction double-channel to generate a cosmetic adverse reaction prediction report.