Cosmetic formulation recommendation method and system based on multi-dimensional features and dynamic calculation

Abstract

This invention relates to a cosmetic formula recommendation method and system based on multi-dimensional features and dynamic estimation. The method includes: receiving structured user requirements and converting them into feature vectors; constructing a formula database and an ingredient efficacy database; performing multi-dimensional similarity matching calculations between the feature vectors and formula feature vectors to obtain a comprehensive similarity score; extracting several formulas with the highest similarity scores as candidate formulas; extracting common components and performing a secondary comprehensive score calculation; generating an optimized ingredient ranking sequence based on descending scores; and calculating the proportion of each ingredient using a benchmark proportion template retrieved from the target product dosage form to generate a recommended cosmetic formula. This invention eliminates requirement ambiguity through six-dimensional structured input, introduces ingredient ranking attenuation weights to achieve accurate matching, and dynamically estimates ingredient proportions using a benchmark proportion template for the dosage form, significantly improving formula development efficiency and success rate while reducing R&D costs.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence, and in particular to a method and system for recommending cosmetic formulas based on multidimensional features and dynamic calculation. Background Technology

[0002] Cosmetic formulation development is the core of cosmetic research and development, directly determining the core efficacy and market value of a product. In the traditional cosmetic research and development process, the formulation development work relies heavily on the personal experience of researchers. This traditional research and development model is no longer suitable for the current market's requirements for the speed and accuracy of product innovation.

[0003] In the traditional cosmetics research and development process, the entire formula development process is based on the personal experience of the R&D personnel. The communication of R&D needs relies heavily on non-standardized written descriptions. The formula screening process is entirely done manually by the R&D personnel. The relevant professional knowledge of formula development also relies on the personal experience of the R&D personnel for inheritance. Newcomers need to accumulate experience through long-term practice before they can participate in formula development work.

[0004] However, the aforementioned traditional manual R&D model has significant industry pain points: First, vague requirements and high communication costs. R&D requirements often rely on non-standardized written descriptions, leading to misunderstandings and repeated communication, resulting in final products that often deviate significantly from the original intentions. Second, inefficient formula screening and difficulty in knowledge transfer. Manual formula screening typically takes 3-5 days and is highly dependent on personal experience, resulting in a training cycle of up to 3 years for new employees. Third, extremely high trial-and-error costs. Due to unclear requirements and limitations in human experience, repeated development is frequent, with a formula development failure rate as high as 40%, and the cost of a single failure reaching 100,000 to 500,000 yuan. Summary of the Invention

[0005] In view of this, it is necessary to provide a cosmetic formulation recommendation method based on multidimensional features and dynamic calculation to address the aforementioned shortcomings of the prior art.

[0006] To address the aforementioned problems, in a first aspect, embodiments of the present invention provide a cosmetic formulation recommendation method based on multidimensional features and dynamic calculation, comprising: Receive the user's structured requirements, transform the structured requirements into feature vectors, and construct a formula database for storing formula data and an ingredient efficacy database for storing ingredient efficacy quantification data. The feature vector is matched with the feature vector of the formula in the formula database to calculate the comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating the efficacy matching degree, the component ranking attenuation weight is introduced so that the higher the ranking of the component in the formula, the higher the corresponding efficacy contribution weight. Several formulas with the highest similarity scores are extracted as candidate formulas. Common components in the candidate formulas are extracted, and a second comprehensive score is calculated for the common components using the ingredient efficacy database. An optimized ingredient ranking sequence is generated based on the comprehensive score in descending order. The corresponding benchmark proportion template is retrieved based on the ingredient ranking sequence and the target product dosage form to estimate the proportion of each ingredient and generate a recommended cosmetic formula.

[0007] Furthermore, cosmetic formulation recommendation methods based on multidimensional features and dynamic inference also include: The generated recommended cosmetic formulas are input into the regulatory engine for compliance verification. Once the verification is successful, a compliant recommended cosmetic formula is output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients.

[0008] Furthermore, the process of receiving the user's structured requirements and converting those requirements into a feature vector includes: The six-dimensional intelligent navigation input layer receives the user's structured needs, which includes a production plant layer, a site of action layer, a product dosage form layer, a usage method layer, an efficacy claim layer, and a user group layer. The structured requirements are transformed into feature vectors through a vectorization processing engine. The feature vectors include a basic framework vector, an efficacy weight vector, and a population vector. The basic framework vector is obtained by uniquely encoding the qualifications of the production plant, the site of action, and the product dosage form. The efficacy weight vector is extracted based on the main efficacy, secondary efficacy, and their weights. The population vector is extracted based on the characteristics of the user population. The formula database stores basic information, ingredient ranking sequence, efficacy labels, and target audience labels for historical compliant formulas and formulas currently on the market; the ingredient efficacy database stores quantitative data on ingredient efficacy, including the intended use of the ingredients, efficacy intensity matrix, and typical concentration range.

[0009] Furthermore, the feature vector is compared with the feature vectors in the formula database using multi-dimensional similarity matching to obtain a comprehensive similarity score, including: Calculate the basic framework similarity, efficacy matching degree, and population matching degree between the feature vector and the formula feature vector; wherein, when calculating the efficacy matching degree, an ingredient ranking attenuation weight is introduced according to the ranking of each ingredient in the formula, so that the ingredient with the higher ranking has a higher efficacy contribution weight. The ingredient ranking attenuation weight is combined with the efficacy intensity data of the ingredient to calculate the individual efficacy matching degree. The overall efficacy matching degree is obtained by summing the individual efficacy matching degrees of all ingredients. Preset weights are assigned to the basic framework similarity, efficacy matching, and population matching. The similarity of each dimension is multiplied by the corresponding preset weight and then summed to obtain the comprehensive similarity score.

[0010] Furthermore, the comprehensive similarity score is calculated using the following formula: Score total =w1×Sim base +w2×Sim eff +w3×Sim aud In the formula, Score total To calculate the overall similarity score, Sim base Based on framework similarity, Sim eff For efficacy matching, Sim aud For audience matching, w1 is the basic framework similarity weight, w2 is the efficacy matching weight, and w3 is the audience matching weight. Among them, the efficacy matching degree Sim eff The calculation introduces a component ranking attenuation weight, which is calculated using the following formula:

[0011] In the formula, Weight rank Weight is the attenuation weight based on ingredient ranking; Rank is the ingredient's position in the formula, with higher rankings resulting in lower Rank values. rank The larger; The efficacy matching degree Sim eff The calculation method is as follows: For each ingredient in the cosmetic formula, the ingredient efficacy intensity corresponding to this ingredient in the ingredient efficacy database is multiplied by the ranking attenuation weight of this ingredient to obtain the single efficacy score of the ingredient. The efficacy matching degree Sim is obtained by summing the single efficacy scores of all ingredients. eff .

[0012] Furthermore, the process involves extracting several formulas with the highest similarity scores as candidate formulas, extracting common components from these candidate formulas, performing a secondary comprehensive score calculation on these common components using the component efficacy database, and generating an optimized component ranking sequence based on the comprehensive score in descending order. This includes: The comprehensive similarity scores are sorted from high to low, and the top-ranked formulas are extracted to form candidate formulas. The common components in the candidate formulas are then extracted. Combining preset efficacy weights, and using quantitative data of ingredients in the ingredient efficacy database, a secondary comprehensive score is calculated for the common ingredients. The secondary comprehensive score includes efficacy matching score, population suitability score, and dosage form suitability score. The common components are arranged in descending order of their secondary comprehensive scores to generate an optimized component ranking sequence.

[0013] Furthermore, the step of retrieving the corresponding baseline proportion template based on the ingredient ranking sequence and the target product dosage form, calculating the proportion of each ingredient, and generating a recommended cosmetic formula includes: Based on the target product dosage form, retrieve the pre-stored benchmark proportion template corresponding to the dosage form, wherein the benchmark proportion template contains the benchmark proportion range corresponding to different ingredient rankings; Determine whether the ingredient efficacy database stores the typical concentration range of the corresponding ingredient; If the ingredient efficacy database stores the typical concentration range of the corresponding ingredient, and the typical concentration range falls within the benchmark proportion range corresponding to the ingredient's ranking, then the precise typical concentration value of the ingredient will be output. If the typical concentration range of the corresponding ingredient is not stored in the ingredient efficacy database, or if the typical concentration range exceeds the benchmark percentage range corresponding to the ranking of the ingredient, then the benchmark percentage range corresponding to the ranking of the ingredient will be output. By integrating the order and proportion of all ingredients, a recommended cosmetic formula is generated.

[0014] Secondly, embodiments of the present invention provide a cosmetic formula recommendation system based on multidimensional features and dynamic calculation, comprising: The six-dimensional intelligent navigation input layer is used to receive users' six-dimensional structured requirements. The six-dimensional intelligent navigation input layer includes a production plant layer, a site of action layer, a product dosage form layer, a usage method layer, an efficacy claim layer, and a user group layer. A vectorization processing layer is used to transform the six-dimensional structured requirements into feature vectors; The core recommendation engine layer has a built-in formula database and ingredient efficacy database. It is used to perform multi-dimensional similarity matching calculations between the feature vector and the formula feature vector in the formula database to obtain a comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating efficacy matching degree, an ingredient ranking attenuation weight is introduced so that the higher the ranking of an ingredient in the formula, the higher the corresponding efficacy contribution weight. The formula generation and estimation layer is used to extract several formulas with the highest similarity scores as candidate formulas, extract common components in the candidate formulas, perform a second comprehensive score calculation on the common components using the component efficacy database, generate an optimized component ranking sequence based on the comprehensive score in descending order, and retrieve the corresponding benchmark proportion template based on the component ranking sequence and the target product dosage form to estimate the proportion of each component and generate a recommended cosmetic formula. The output and compliance layer is used to input the generated cosmetic recommendation formulas into the regulatory engine for compliance verification. After the verification is passed, the compliant cosmetic recommendation formulas are output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients.

[0015] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the cosmetic formulation recommendation method based on multidimensional features and dynamic calculation as described in the first aspect embodiment of the present invention.

[0016] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps in the cosmetic formulation recommendation method based on multidimensional features and dynamic calculation as described in the first aspect embodiment of the present invention.

[0017] The cosmetic formulation recommendation method and system based on multidimensional features and dynamic calculation provided by this invention have the following advantages compared with the prior art: (1) This invention transforms user requirements into structured data and performs feature vectorization processing through a six-dimensional intelligent navigation input layer, eliminating the ambiguity and communication loss caused by natural language descriptions in traditional R&D. It reduces the requirement communication time from the traditional 2 hours to 5 minutes, achieving accurate expression and efficient transmission of R&D requirements, improving the efficiency of requirement communication, and shortening the formula development cycle. The multi-dimensional similarity matching algorithm reduces formula screening from 3 days of manual processing to AI-level response in seconds, shortening the overall development cycle by nearly 75%, and realizing full automation from requirement input to compliant formula generation.

[0018] (2) This invention introduces a component ranking attenuation weight in the efficacy matching calculation, making the components with higher rankings contribute more to the overall efficacy. Combined with the quantitative calibration data of the component efficacy database, it achieves accurate quantitative matching of the formula. By using a regulatory engine to perform compliance verification on the generated formula, the compliance and security of the formula are effectively guaranteed, and the trial and error cost and failure risk of formula development are significantly reduced. The success rate of formula market is increased from the traditional 40% to 70%, and the trial and error cost of a single formula is reduced by 500,000 yuan, which greatly saves the company's annual R&D expenses.

[0019] (3) This invention extracts common components from candidate formulations and performs secondary comprehensive score calculations. Combined with dosage form benchmark ratio templates, it dynamically calculates the proportion of each component, resulting in formulations with strong engineering application guidance. At the same time, a feedback optimization mechanism is introduced, enabling the system to continuously self-correct and constructing a cosmetic formulation decision-making knowledge base with self-evolution capabilities. Attached Figure Description

[0020] Figure 1 A flowchart of the cosmetic formula recommendation method based on multidimensional features and dynamic calculation provided by the present invention; Figure 2 This is an architecture diagram of the six-dimensional intelligent navigation input layer provided by the present invention; Figure 3 A schematic diagram of the compliance verification system of the regulatory engine provided by the present invention; Figure 4 The overall architecture diagram of the cosmetic formula recommendation system based on multidimensional features and dynamic inference provided by the present invention; Figure 5 This is a structural block diagram of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0023] Before describing the specific implementation methods, the core technical terms involved in this invention will be defined and explained first: The six-dimensional intelligent navigation input layer refers to the input interface constructed by this invention for receiving users' structured needs. It includes six dimensions: production plant layer, site of action layer, product dosage form layer, usage method layer, efficacy claim layer, and user group layer. These dimensions together constitute a complete formula requirement description system.

[0024] A vectorization processing engine is a processing module that transforms discrete, structured requirement data into machine-computable feature vectors. It achieves the numerical transformation of data through methods such as one-hot encoding and weight extraction.

[0025] One-hot encoding is a method of encoding categorical variables into binary vectors, where each category corresponds to an independent binary bit. A bit that is 1 indicates that the variable belongs to that category, and a bit that is 0 indicates that the variable does not belong to that category.

[0026] Feature vector: A one-dimensional array composed of feature values ​​of multiple dimensions. It is a carrier of demand information that can be recognized and calculated by machines. The feature vector in this invention consists of three core sub-vectors: basic framework vector, efficacy weight vector, and population vector.

[0027] Basic framework vector: One of the core sub-vectors of the feature vector, used to carry the basic framework class requirement information of formula development, and realizes the vectorization transformation of classification features through one-hot encoding.

[0028] Efficacy weight vector: One of the core sub-vectors of the feature vector, used to carry efficacy-related requirement information for formula development, including the primary efficacy, secondary efficacy, and numerical weight of each efficacy.

[0029] Audience Vector: One of the core sub-vectors of the feature vector, used to carry the target customer group's needs information for formula development, and to extract and quantify the relevant features of the target user group.

[0030] The formula database is a data warehouse used to store historical compliant formulas and formulas currently on the market. It contains core data such as basic formula information, ingredient ranking sequence, efficacy labels, and target audience labels.

[0031] The ingredient efficacy database is a data warehouse used to store quantitative calibration data of all legal cosmetic ingredients, including quantitative information such as the purpose of use, efficacy intensity matrix, and typical concentration range of the ingredients.

[0032] Efficacy intensity refers to the quantitative value of an ingredient in a specific efficacy dimension, which is pre-calibrated in the ingredient efficacy database. For example, the repair intensity of ceramide is 0.9 and the moisturizing intensity is 0.7.

[0033] Typical concentration range refers to the concentration range of an ingredient that is typically used in compliant formulations, and is used to guide the calculation of its proportion when generating a formulation.

[0034] Basic framework similarity refers to the degree of similarity between the basic framework vector in the user demand feature vector and the basic framework vector of the formula in the formula database. It is used to measure the matching of basic dimensions such as the qualifications of the production plant, the site of action, and the product dosage form.

[0035] Efficacy matching degree refers to the degree of similarity between the efficacy weight vector in the user demand feature vector and the efficacy features of the formula in the formula database. It is used to measure the matching of the main efficacy, secondary efficacy and their weight with the actual efficacy performance of the formula.

[0036] Audience matching degree refers to the similarity between the audience vector in the user demand feature vector and the audience tags of the formula in the formula database. It is used to measure the matching of the target customer group characteristics with the applicable audience of the formula.

[0037] The overall similarity score is a comprehensive score obtained by weighting and summing the basic framework similarity, efficacy matching degree, and population matching degree. It is used to characterize the overall matching degree between user needs and formulas in the formula library.

[0038] Ingredient ranking attenuation weight refers to the weight value assigned to an ingredient based on its ranking number in the formula. Ingredients ranked higher have a greater weight, which is used to characterize the differentiated impact of ingredients with different rankings on the overall efficacy of the formula.

[0039] The individual efficacy score refers to the actual efficacy contribution value of a single component after considering the ranking attenuation weight, which is obtained by multiplying the efficacy intensity of the component by its ranking attenuation weight.

[0040] The regulatory engine refers to an automated verification module with built-in multi-dimensional compliance verification rules, used to perform compliance checks on the generated recommended cosmetic formulas to ensure that the formulas meet relevant regulatory requirements.

[0041] The user feedback evolution engine is a closed-loop optimization module used to collect formula trial feedback data and feed it back into the recommendation model. It updates the weight parameters of the similarity algorithm through multiple rounds of granular scoring data to achieve continuous evolution of formula recommendations.

[0042] Figure 1 This is a flowchart illustrating the cosmetic formulation recommendation method based on multidimensional features and dynamic estimation provided by the present invention. (Refer to...) Figure 1 The method includes the following steps: Step S1: Receive the user's structured requirements, convert the structured requirements into feature vectors, and construct a formula database for storing formula data and a component efficacy database for storing component efficacy quantification data.

[0043] Specifically, to eliminate the technical problems caused by non-standardized textual descriptions in traditional cosmetic formulation development, such as misunderstandings of requirements, significant communication losses, low efficiency in formula screening, and difficulties in knowledge transfer, this invention constructs a six-dimensional structured requirement input mechanism. This mechanism eliminates ambiguity in non-standardized requirements from the source, transforms R&D requirements into structured data that can be recognized and calculated by machines, and builds a formula database and an ingredient efficacy database, thus establishing a complete underlying data architecture for subsequent intelligent formula matching and generation.

[0044] In some embodiments of this application, step S1 includes input structuring and feature vectorization processing based on six-dimensional intelligent navigation, as well as a corresponding high-dimensional database construction step, specifically including the following sub-steps S11~S13: S11 receives users' structured requests through a six-dimensional intelligent navigation input layer.

[0045] Figure 2 This is an architecture diagram of the six-dimensional intelligent navigation input layer provided by the present invention. Figure 2 As shown, the six-dimensional intelligent navigation input layer specifically includes six dimensions: manufacturing plant layer, site of action layer, product dosage form layer, usage method layer, efficacy claim layer, and target user layer. The manufacturing plant layer describes manufacturing capabilities and process requirements, including specific factory qualification information such as Marubi Biotechnology Co., Ltd. and Shengmei Cosmetics Co., Ltd. This dimension provides production feasibility constraints for subsequent formula recommendations. The site of action layer describes the application area of ​​the product, including specific areas such as hair, body hair, face, eyes, lips, and nails. This dimension determines the applicability requirements of the formula. The product dosage form layer describes the product's technical framework and physical form, including specific dosage forms such as creams, serums, cleansers, and eyeshadows. This dimension provides a basis for selecting dosage form benchmark templates for subsequent proportion calculations. The usage method layer describes the product's usage scenarios, including two main categories: leave-on products and rinse-off products. This dimension affects the safety and user experience requirements of the formula. The efficacy claim layer describes the product's core value proposition, including specific effects such as moisturizing, anti-wrinkle, and firming, and includes primary and secondary effects and their corresponding weight allocation. This dimension is the core basis for calculating efficacy matching degree. The target audience layer describes the characteristics of the product's target customer group, including specific demographic attributes such as skin type, age, gender, special needs, skincare concerns, and lifestyle. This dimension determines the targeted adaptation requirements of the formula.

[0046] Users can input their needs through the interactive interface provided by the six-dimensional intelligent navigation input layer, according to the six preset dimensions. For example, in the production plant layer, users can select specific factory qualification requirements; in the application site layer, they can select the target application site; in the product dosage form layer, they can select the desired product form; in the application method layer, they can select the leave-in or rinse-off type; in the efficacy claim layer, they can set the main efficacy, secondary efficacy and their corresponding weight values; and in the target audience layer, they can select the skin type, age range, gender, special needs, skin care demands and lifestyle characteristics of the target customer group.

[0047] S12, the structured requirements are transformed into feature vectors through a vectorization processing engine.

[0048] The feature vector includes a basic framework vector, an efficacy weight vector, and a population vector. The basic framework vector is obtained by uniquely encoding the manufacturer's qualifications, the site of action, and the product dosage form. The efficacy weight vector is extracted based on the primary efficacy, secondary efficacy, and their weights. The population vector is extracted based on the characteristics of the user population.

[0049] Specifically, the basic framework vector is generated by performing one-hot encoding on three discrete categorical data points: manufacturing plant qualifications, site of action, and product dosage form. One-hot encoding involves converting each categorical variable into a binary vector where only one bit is 1 and the rest are 0, representing a specific category value. For example, for the site of action dimension, the face, eyes, and lips each correspond to different binary encoding bits. When a user selects the face, the encoding bit for the face is 1, while the encoding bits for the other parts are all 0.

[0050] The efficacy weight vector is generated as follows: primary efficacy, secondary efficacy, and their corresponding weight values ​​are extracted from the efficacy claim layer input by the user, and this information is combined to form the efficacy weight vector. For example, if the user sets the primary efficacy to moisturizing with a weight of 0.6 and the secondary efficacy to anti-wrinkle with a weight of 0.4, then the corresponding position in the efficacy weight vector for moisturizing will have a value of 0.6, and the corresponding position for anti-wrinkle will have a value of 0.4.

[0051] The crowd vector is generated by extracting target customer features from the user input user group layer, including skin type (sensitive or dry skin), age range, gender, special needs (e.g., suitable for pregnant women), skin care needs (e.g., whitening), lifestyle (e.g., people who stay up late), and combining these features to form the crowd vector.

[0052] S13, pre-build formula database and ingredient efficacy database.

[0053] The formula database stores data related to historical compliant formulas and formulas currently on the market. Specific stored content includes: basic formula information such as formula name and registration number; ingredient ranking sequence (arranged in descending order of quantity, with trace amounts removed to ensure representativeness of core ingredients); efficacy labels such as moisturizing, anti-wrinkle, and repairing; and target skin labels such as suitable for sensitive skin or suitable for mature skin.

[0054] The ingredient efficacy database stores quantitatively calibrated data for all legally approved cosmetic ingredients. Specific stored content includes: the intended use of the ingredient (e.g., moisturizer, emulsifier, preservative); an efficacy intensity matrix, which is a data set that quantitatively calibrates the efficacy intensity of each ingredient across different efficacy dimensions; for example, ceramides are calibrated to a repair intensity of 0.9 and a moisturizing intensity of 0.7; and a typical concentration range, which is the concentration range typically used for the ingredient in compliant formulations; for example, the typical concentration range for a certain ingredient is 1%-5%.

[0055] This invention transforms user needs into structured data and performs feature vectorization processing through a six-dimensional intelligent navigation input layer. This eliminates the ambiguity and communication loss caused by natural language descriptions in traditional R&D, compresses the communication time for needs from the traditional 2 hours to 5 minutes, and achieves accurate expression and efficient transmission of R&D needs, thereby improving the efficiency of needs communication and shortening the formula development cycle.

[0056] Step S2: Perform multi-dimensional similarity matching calculation between the feature vector and the formula feature vector in the formula database to obtain a comprehensive similarity score; wherein, the multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating efficacy matching degree, an ingredient ranking attenuation weight is introduced so that the higher the ranking of an ingredient in the formula, the higher the corresponding efficacy contribution weight.

[0057] Understandably, manual screening in traditional cosmetic formulation selection is time-consuming, and ordinary matching algorithms, which only focus on the presence or absence of ingredients while ignoring the industry characteristic that ingredient ranking determines the actual contribution to efficacy, lead to significant matching biases. To address this issue, this invention sets up a multi-dimensional similarity matching algorithm, performing weighted matching from three dimensions: basic framework, efficacy performance, and target audience suitability. Furthermore, it innovatively introduces an ingredient ranking attenuation weight in the efficacy matching calculation, allowing top-ranked core ingredients to contribute more significantly to the efficacy matching score. This achieves precise quantitative matching between user needs and existing formulations, efficiently screening candidate formulations that meet R&D requirements and providing a high-quality candidate formulation foundation for subsequent formulation generation.

[0058] In the multi-dimensional similarity matching calculation in step S2, it is necessary to calculate the basic framework similarity, efficacy matching degree, and population matching degree. In some embodiments of this application, step S2 specifically includes the following sub-steps S21 to S24: S21, Calculate the basic framework similarity between the feature vector and the recipe feature vector.

[0059] Basic framework similarity Sim baseThe calculation is based on the basic framework vector generated in step S1 and the basic framework vector pre-stored in the formula database. The basic framework vector is obtained by uniquely encoding the qualifications of the production plant, the site of action, and the product dosage form. Therefore, the calculation of basic framework similarity adopts the cosine similarity algorithm, which calculates the cosine value of the angle between two vectors. The closer the cosine value is to 1, the more similar the two vectors are, that is, the higher the degree of basic framework dimension matching.

[0060] S22, calculate the population matching degree between the feature vector and the formula feature vector.

[0061] Audience Matching Sim aud The calculation is based on the population vector generated in step S1 and the population tag vector pre-stored in the formula database. The population vector is extracted based on the characteristics of the user population, including features such as skin type, age range, gender, special needs, skincare demands, and lifestyle. The population matching degree is also calculated using the cosine similarity algorithm, and the obtained population matching degree value is used to characterize the degree of fit of the target customer group's characteristics.

[0062] S23, calculate the efficacy matching degree between the feature vector and the formula feature vector.

[0063] First, obtain the ranking sequence of each ingredient in the formula. The formula database stores formula information containing an ingredient ranking sequence, which is arranged from high to low according to the amount of ingredients added in the formula. Rank 1 indicates the first ingredient, i.e., the ingredient added in the highest amount, Rank 2 indicates the second ingredient, and so on.

[0064] The ranking attenuation weight of each component is calculated based on its ranking sequence number. This invention innovatively introduces a formula for calculating the component ranking attenuation weight:

[0065] In the formula, Weight rank This is the weighting factor for component ranking attenuation, where Rank is the component's position in the formulation. According to this formula, the earlier a component ranks, the smaller its Rank value. The calculated Weight... rank The larger the value, the higher the weight of that component in terms of its contribution to the overall efficacy. For example, the component ranked first (Rank=1) has a higher weight for attenuation. rank =1 / =1; the fourth-ranked component (Rank=4) has a decay weight of Weight. rank =1 / =0.5. This design reflects the dominant role of the core ingredients in the formula in the product's efficacy and aligns with professional understanding of cosmetic formulation.

[0066] Next, the efficacy intensity data for each ingredient is retrieved from the ingredient efficacy database. This database stores an efficacy intensity matrix that quantifies all legally approved ingredients; for example, ceramides are calibrated for a repair intensity of 0.9 and a moisturizing intensity of 0.7; hyaluronic acid is calibrated for a moisturizing intensity of 0.95 and a repair intensity of 0.3. For each ingredient, the corresponding efficacy intensity value is extracted based on the user's desired efficacy dimension.

[0067] Further, calculate the individual efficacy score for each component. Multiply the efficacy intensity of each component by its corresponding ranking attenuation weight to obtain the individual efficacy score for that component. The calculation formula is: Individual Efficacy Score = Component Efficacy Intensity × Weight rank .

[0068] Finally, the individual efficacy scores of all ingredients are summarized to obtain the efficacy matching degree Sim. eff That is, for all ingredients in a formula, their individual efficacy scores are summed, and the result is the efficacy match between the formula and the user's needs.

[0069] S24. Calculate the overall similarity score based on the basic framework similarity, efficacy matching degree, and population matching degree.

[0070] Specifically, preset weights are assigned to the basic framework similarity, efficacy matching, and population matching. In a preferred embodiment of the invention, the basic framework similarity weight w1 is set to 0.3, the efficacy matching weight w2 is set to 0.5, and the population matching weight w3 is set to 0.2. These weight values ​​are set based on professional considerations of cosmetic formulation development practice: efficacy is the core value of a formulation, therefore it is given the highest weight of 0.5, which aligns with the core needs of cosmetic formulation development.

[0071] The overall similarity score (Score) is obtained by multiplying the similarity scores of each dimension by their corresponding preset weights and then summing the results. total The overall similarity score is calculated using the following formula: Score total =w1×Sim base +w2×Sim eff +w3×Sim aud In the formula, Score total To calculate the overall similarity score, Sim base Based on framework similarity, Sim eff For efficacy matching, Sim aud For audience matching, w1 is the basic framework similarity weight, w2 is the efficacy matching weight, and w3 is the audience matching weight.

[0072] Through the above scheme, this invention generates a comprehensive similarity score for each recipe in the recipe database. This score quantifies the overall degree of matching between the recipe and user needs. The higher the comprehensive similarity score, the better the match between the recipe and user needs. Candidate recipes will then be selected based on this score.

[0073] This invention achieves comprehensive and accurate matching between user R&D needs and existing formulas through multi-dimensional cosine similarity and weighted matching calculations, breaking through the limitations of traditional algorithms that only focus on the presence or absence of ingredients. This invention introduces ingredient ranking attenuation weights in efficacy matching calculations, making higher-ranking ingredients contribute more to overall efficacy. Combined with quantitative calibration data from an ingredient efficacy database, it achieves precise quantitative matching of formulas. Compliance verification of generated formulas through a regulatory engine effectively ensures the compliance and security of formulas, significantly reducing the trial-and-error costs and failure risks in formula development. It increases the formula market success rate from the traditional 40% to 70%, reduces the trial-and-error cost of a single formula by 500,000 yuan, and greatly saves companies' annual R&D expenses.

[0074] Step S3: Extract several formulas with the highest similarity scores as candidate formulas, extract common components from the candidate formulas, perform a second comprehensive score calculation on the common components using the component efficacy database, generate an optimized component ranking sequence based on the comprehensive score in descending order, and retrieve the corresponding benchmark proportion template based on the component ranking sequence and the target product dosage form to estimate the proportion of each component and generate a recommended cosmetic formula.

[0075] S31, extract several formulas with the highest similarity scores as candidate formulas, and extract the common components in the candidate formulas.

[0076] First, the overall similarity score calculated in step S2 is obtained. total All recipes in the recipe database are sorted from highest to lowest score. A higher overall similarity score indicates a better overall match between the recipe and the user's needs.

[0077] Next, several top-ranked recipes are extracted as candidate recipes. In a preferred embodiment of the invention, the top K recipes in terms of similarity scores are extracted as candidate recipes. The value of K can be adjusted according to the size of the recipe library and the actual application scenario. For example, K can be set to ten or twenty to ensure that the candidate recipes are sufficiently representative while avoiding the introduction of too much noisy data.

[0078] Next, component analysis is performed on the extracted candidate formulations to extract common components shared across all candidate formulations. Common components refer to the set of components that appear in all candidate formulations; these components have been validated by multiple successful formulations and possess high reliability and reference value. The extraction of common components is based on the following technical understanding: components that appear in multiple successful formulations highly aligned with user needs are often the core foundational components for achieving the target efficacy, possessing high reliability and reference value. For example, if all ten candidate formulations contain ceramides and hyaluronic acid, then ceramides and hyaluronic acid are extracted as common components. The extraction of common components overcomes the limitations of single formulations, integrates the advantages of multiple excellent formulations, and provides a high-quality component foundation for generating optimized formulations.

[0079] S32, combining preset efficacy weights, using the quantitative data of ingredients in the ingredient efficacy database, perform a second comprehensive score calculation on the common ingredients, and generate an optimized ingredient ranking sequence in descending order based on the comprehensive score.

[0080] Specifically, for each component in the set of common components extracted in the above steps, its corresponding quantitative data is obtained from the component efficacy database. The component efficacy database stores efficacy intensity matrices, population suitability information, and dosage form suitability information for all legal components that have been quantitatively calibrated.

[0081] A secondary comprehensive score is calculated for each common ingredient, taking into account the efficacy weighting of user needs, population characteristics, and product dosage form. The secondary comprehensive score refers to the overall score of each ingredient recalculated using quantitative data from an ingredient efficacy database, based on the common ingredients in the candidate formulations and incorporating the efficacy weighting of user needs, population characteristics, and product dosage form. This score is used to generate an optimized ingredient ranking sequence. The secondary comprehensive score includes efficacy matching score, population suitability score, and dosage form suitability score. The efficacy matching score is calculated based on the degree of matching between the ingredient's efficacy intensity and the efficacy weighting of user needs. The population suitability score is calculated based on the degree of matching between the ingredient's population suitability attributes and the characteristics of the user's target population. The dosage form suitability score is calculated based on the compatibility between the ingredient and the target product dosage form.

[0082] The efficacy matching score is calculated as follows: the efficacy intensity matrix of an ingredient in the ingredient efficacy database is matched with the efficacy weight vector of the user's needs to obtain the matching score of the ingredient in the efficacy dimension. For example, if the user's needs have a moisturizing efficacy weight of 0.6 and an anti-wrinkle efficacy weight of 0.4, and ingredient A has a moisturizing intensity of 0.9 and an anti-wrinkle intensity of 0.3, then the efficacy matching score of ingredient A is equal to 0.9 multiplied by 0.6 plus 0.3 multiplied by 0.4, which equals 0.66.

[0083] The audience fit score is calculated by matching the audience fit tags of an ingredient with the characteristics of the user's needs, thus obtaining the fit score of that ingredient in the audience dimension. For example, if the user's need is for people with sensitive skin, and ingredient B is labeled as suitable for sensitive skin, then the audience fit score will be high.

[0084] The dosage form compatibility score is calculated by matching the compatibility of an ingredient with the target product dosage form, thus obtaining the ingredient's compatibility score in the dosage form dimension. For example, if the target dosage form is a serum, and ingredient C is marked as suitable for serum dosage forms in the ingredient efficacy database, then its dosage form compatibility score will be high.

[0085] The efficacy matching score, population suitability score, and dosage form suitability score of the common ingredients are added together to obtain the secondary comprehensive score of the ingredient. The secondary comprehensive score comprehensively measures the degree of overall matching between the ingredient and user needs in the three dimensions of efficacy performance, population suitability, and dosage form suitability.

[0086] Finally, all components in the common component set are sorted in descending order of their secondary comprehensive scores. Components with higher secondary comprehensive scores indicate a better overall match with user needs and should occupy a higher position in the final formula. Through this sorting process, this invention generates an optimized component ranking sequence. This ranking sequence differs from the ranking order of any single historical formula; instead, it is a completely new ranking that integrates the advantages of multiple excellent formulas and is dynamically optimized for current user needs, reflecting the formula innovation and personalized adaptability of this invention.

[0087] S33. Based on the ingredient ranking sequence and the target product dosage form, retrieve the corresponding benchmark proportion template, calculate the proportion of each ingredient, and generate a recommended cosmetic formula.

[0088] The benchmark proportion template refers to a technical template pre-set for different product dosage forms, specifying the benchmark proportion range corresponding to different ranking intervals. The typical concentration range refers to the concentration range of the ingredient typically used in compliant formulations, stored in the ingredient efficacy database. The precise typical concentration value refers to the specific concentration value of the ingredient directly output when its typical concentration range falls within the benchmark proportion range. The benchmark proportion range refers to the proportion range of the ingredient based on its ranking provided by the benchmark proportion template when a precise typical concentration value cannot be output.

[0089] Specifically, based on the target product dosage form input by the user, a pre-stored benchmark percentage template corresponding to that dosage form is retrieved. This invention pre-establishes differentiated benchmark percentage templates for different product dosage forms, such as templates for serums, creams, lotions, and cleansers. Each benchmark percentage template includes benchmark percentage ranges corresponding to different ingredient rankings. Taking a serum dosage form as an example, the benchmark percentage for the first to third ranked ingredients is set at 15% to 30%, the benchmark percentage for the fourth to sixth ranked ingredients is set at 5% to 15%, and the benchmark percentage for the seventh ranked ingredients and beyond is set at less than 5%. These benchmark percentage ranges are based on professional knowledge of cosmetic formulation and statistical analysis of a large amount of actual formulation data.

[0090] For each component in the optimized component ranking sequence, the corresponding baseline proportion range is determined according to its ranking order. Simultaneously, the typical concentration range for that component is stored in the component efficacy database.

[0091] Decision-making process for percentage estimation: If the ingredient efficacy database stores the typical concentration range of the ingredient, and this typical concentration range falls entirely within the baseline percentage range corresponding to the ingredient's ranking, then the precise typical concentration value of the ingredient is output. The output of the precise typical concentration value means that the amount of the ingredient added in the recommended formulation can directly adopt the validated typical concentration, ensuring both the effectiveness and reliability of the formulation.

[0092] If the ingredient efficacy database does not store the typical concentration range of an ingredient, or if the stored typical concentration range exceeds the baseline proportion range corresponding to the ingredient's ranking, then the baseline proportion range corresponding to the ingredient's ranking will be output. The output of the baseline proportion range implies that the amount of this ingredient added needs to be further determined through experiments during subsequent research and development. The currently recommended formulation provides a reasonable concentration range as a starting point for research and development.

[0093] For example, for the top-ranked ingredient, ceramide, its corresponding baseline proportion range is 15% to 30%. The typical concentration range of ceramide stored in the ingredient efficacy database is 20% to 25%, which falls within the baseline proportion range. Therefore, the output is a precise typical concentration value of 20% to 25%. For the fifth-ranked ingredient, xanthan gum, its corresponding baseline proportion range is 5% to 15%, but the typical concentration range of xanthan gum is not stored in the ingredient efficacy database. Therefore, the output is a baseline proportion range of 5% to 15%.

[0094] Finally, by integrating the order of all ingredients and their respective proportions, a complete recommended cosmetic formula is generated. This recommended formula includes the ingredient order and the corresponding proportion of each ingredient. For some ingredients, precise typical concentration values ​​are provided, while for others, baseline proportion ranges are provided. Together, these constitute a research and development formula scheme that is both clearly instructive and allows for reasonable adjustments.

[0095] This embodiment extracts common components from candidate formulations and screens out core effective ingredients that have been validated in the market, ensuring the effectiveness and safety of the optimized formulation. Through three-dimensional secondary comprehensive score calculation, intelligent optimization of ingredient ranking is achieved, generating a new formulation framework fully adapted to user needs. Based on the dynamic proportion calculation of the dosage form benchmark proportion template, standardized and precise setting of ingredient addition amounts is achieved, eliminating the reliance on experience in traditional manual settings, improving the engineering feasibility of the formulation, shortening the formulation optimization cycle, and providing a feasible intelligent generation solution for cosmetic formulation development.

[0096] In some embodiments of this application, the cosmetic formulation recommendation method based on multidimensional features and dynamic estimation further includes: Step S4: Input the generated recommended cosmetic formula into the regulatory engine for compliance verification. After the verification is passed, the compliant recommended cosmetic formula is output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients.

[0097] First, the recommended cosmetic formula generated in step S3 is input into the regulatory engine. This recommended formula includes the ingredient ranking sequence and the proportion of each ingredient, with some ingredients showing precise typical concentration values ​​and others showing baseline proportion ranges. The regulatory engine initiates a compliance verification process, executing the following verification rules for each ingredient in the formula: The first verification rule is the verification of the highest historical usage limit of the used cosmetic ingredients catalog. The regulatory engine has built-in complete data of the used cosmetic ingredients catalog. For each ingredient in the formula, it queries the highest historical usage value recorded in the catalog. If the recommended percentage of the ingredient exceeds the highest historical usage value, the verification is deemed unsuccessful, and the formula adjustment process must be triggered; if the recommended percentage does not exceed the highest historical usage value, the verification passes. For ingredients with an output baseline percentage range, the regulatory engine will check whether the upper limit of the range exceeds the highest historical usage limit. If it does, the range needs to be narrowed or the ingredient needs to be replaced.

[0098] The second verification rule is the CIR (Cosmetic Ingredient Review) safe use standard verification. The regulatory engine has built-in CIR-published safe use standards for leave-on and rinse-off products. Based on the user-input usage method information, it determines whether the target product is a leave-on or rinse-off product, and then calls the corresponding safe use standard. For each ingredient in the formula, it queries the upper limit or usage restriction of the safe use concentration specified in the CIR standard. If the recommended proportion of the ingredient meets the CIR standard requirements, the verification passes; otherwise, it fails.

[0099] If all ingredients pass the two compliance checks mentioned above, the regulatory engine will output that the formula is a compliant cosmetic recommendation formula, which can be directly used for subsequent research and development or production. If any ingredient fails the check, the regulatory engine will report non-compliance information and trigger the formula adjustment mechanism, such as suggesting reducing the concentration of the ingredient, replacing it with an alternative ingredient, or regenerating the formula.

[0100] Furthermore, the system incorporates a user feedback evolution engine to construct a closed-loop optimization mechanism for the entire formula recommendation process. The system first puts validated, compliant cosmetic formulas into an internal scoring and trial phase, collecting granular scoring data from internal R&D personnel on dimensions such as formula efficacy, dosage form suitability, and production feasibility. After completing the internal trial phase, the formula is put into the market validation phase, collecting multi-dimensional granular data on user feedback, efficacy evaluation, and market performance. The user feedback evolution engine feeds back the collected multi-round granular scoring data into the model weight parameters of the preceding multi-dimensional similarity matching algorithm, iteratively optimizing the preset weights corresponding to the basic framework similarity, efficacy matching, and target audience matching. Simultaneously, it updates the quantitative calibration data of ingredients in the ingredient efficacy database, enabling continuous evolution of the formula recommendation logic. This transforms the traditional one-way formula generation model into a fully closed-loop evolutionary model encompassing initial recommendation, internal scoring and trial, market validation, and large-scale model optimization.

[0101] Figure 3 This is a schematic diagram of the compliance verification system of the regulatory engine provided by the present invention. Figure 3This diagram presents a multi-dimensional, cross-validation architecture with overlapping rules. The core of the diagram is the regulatory engine, whose primary function is to ensure the compliance and security of the final formulation data. Around the regulatory engine, four compliance validation dimensions are set as inputs: the highest historical usage requirements corresponding to the catalog of used cosmetic ingredients, distinguishing between rinse-off and leave-on products; information on the use of marketed ingredients, specifying the site of action, method of application, and dosage; an index of international cosmetic safety assessment data, specifying the site of action, method of application, and dosage; and ingredient usage standards published by the international authoritative organization CIR, distinguishing between leave-on and rinse-off products. The compliance rules of these four dimensions overlap, forming a multi-dimensional, cross-validation compliance and security boundary. This filters out ingredient and dosage data that do not meet the requirements of any dimension, ultimately outputting fully compliant formulation data, achieving comprehensive compliance control over the formulation.

[0102] This invention constructs a formula compliance assurance system covering both domestic and international standards through fully automated compliance verification using a multi-dimensional regulatory engine. This ensures a high compliance pass rate for generated formulas, mitigating the R&D trial-and-error risks caused by non-compliant raw materials from the source and reducing R&D costs. Simultaneously, a closed-loop optimization mechanism built through a user feedback evolution engine enables continuous self-iteration of the formula recommendation algorithm, allowing the system to continuously adapt to changes in market demand. This overcomes the limitations of traditional static databases, constructing a self-evolving cosmetic formula R&D system and providing the industry with a long-term reusable intelligent R&D solution.

[0103] Figure 4 This is a diagram illustrating the overall architecture of the cosmetic formula recommendation system based on multidimensional features and dynamic calculation provided by the present invention, with reference to... Figure 4 The system includes: The six-dimensional intelligent navigation input layer is used to receive users' six-dimensional structured requirements. The six-dimensional intelligent navigation input layer includes a production plant layer, a site of action layer, a product dosage form layer, a usage method layer, an efficacy claim layer, and a user group layer. A vectorization processing layer is used to transform the six-dimensional structured requirements into feature vectors; The core recommendation engine layer has a built-in formula database and ingredient efficacy database. It is used to perform multi-dimensional similarity matching calculations between the feature vector and the formula feature vector in the formula database to obtain a comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating efficacy matching degree, an ingredient ranking attenuation weight is introduced so that the higher the ranking of an ingredient in the formula, the higher the corresponding efficacy contribution weight. The formula generation and estimation layer is used to extract several formulas with the highest similarity scores as candidate formulas, extract common components in the candidate formulas, perform a second comprehensive score calculation on the common components using the component efficacy database, generate an optimized component ranking sequence based on the comprehensive score in descending order, and retrieve the corresponding benchmark proportion template based on the component ranking sequence and the target product dosage form to estimate the proportion of each component and generate a recommended cosmetic formula. The output and compliance layer is used to input the generated cosmetic recommendation formulas into the regulatory engine for compliance verification. After the verification is passed, the compliant cosmetic recommendation formulas are output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients.

[0104] The cosmetic formula recommendation system based on multidimensional features and dynamic calculation provided by this invention is used to execute the cosmetic formula recommendation method based on multidimensional features and dynamic calculation provided in the foregoing embodiments. The cosmetic formula recommendation method based on multidimensional features and dynamic calculation has been described in detail in the above embodiments, and will not be repeated here.

[0105] like Figure 4 As shown, the system adopts a layered architecture design, which includes, from bottom to top, a six-dimensional intelligent navigation input layer, a vectorization processing layer, a core recommendation engine layer, a recipe generation and calculation layer, and an output and compliance layer. Each layer works together to complete the intelligent processing of the entire process from user demand input to compliant recipe generation.

[0106] The six-dimensional intelligent navigation input layer serves as the user interface for the system. This layer comprises six dimensions: production plant layer, site of action layer, product dosage form layer, usage method layer, efficacy claim layer, and target user group layer. It receives structured demand information from users according to preset dimensions, providing standardized input for subsequent processing.

[0107] The vectorization processing layer is connected to the six-dimensional intelligent navigation input layer and has a built-in input feature vectorization module. This layer transforms the received six-dimensional structured requirements into machine-computable feature vectors, specifically including basic framework vectors, efficacy weight vectors, and population vectors, realizing the transformation from user needs to quantitative data.

[0108] The core recommendation engine layer incorporates a multi-dimensional similarity matching algorithm, which calculates a comprehensive similarity score by evaluating the similarity of the basic framework, efficacy matching, and demographic matching. The efficacy matching calculation incorporates a weighted approach based on ingredient ranking attenuation. The formula database stores historical and market formula data, while the ingredient efficacy database stores the efficacy intensity matrix and typical concentration ranges of the ingredients, providing data support for the matching calculation.

[0109] The formulation generation and calculation layer includes two core modules: a formulation generation engine and a dynamic component ratio calculation system. The former is responsible for extracting common components of candidate formulations and performing secondary comprehensive score calculations to generate an optimized ranking sequence, while the latter dynamically calculates the ratio of each component based on the ranking sequence and the target product dosage form by retrieving the benchmark ratio template.

[0110] The output and compliance layer is connected to the formula generation and calculation layer. This layer has a built-in regulatory engine filtering module that performs compliance verification on the generated recommended formulas. The verification rules include the highest historical usage limit of the catalog of used cosmetic ingredients and CIR safe usage standards. After the verification is passed, the final compliant cosmetic recommended formula and dynamic proportion suggestions are output, completing the intelligent formula recommendation process.

[0111] Figure 5 The structural block diagram of the electronic device provided by the present invention is as follows: Figure 5 As shown, the present invention also provides an electronic device 500, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The electronic device 500 includes a processor 501 and a memory 502, wherein the memory 502 stores a cosmetic formula recommendation program 503 based on multi-dimensional features and dynamic calculation.

[0112] In some embodiments, memory 502 may be an internal storage unit of a computer device, such as a hard disk or memory. In other embodiments, memory 502 may be an external storage device of a computer device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, memory 502 may include both internal and external storage units of the computer device. Memory 502 is used to store application software and various types of data installed on the computer device, such as program code for installing the computer device. Memory 502 can also be used to temporarily store data that has been output or will be output. In one embodiment, when the cosmetic formula recommendation program 503 based on multi-dimensional features and dynamic calculation is executed by processor 501, the following steps are implemented: Receive the user's structured requirements, transform the structured requirements into feature vectors, and construct a formula database for storing formula data and an ingredient efficacy database for storing ingredient efficacy quantification data. The feature vector is matched with the feature vector of the formula in the formula database to calculate the comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating the efficacy matching degree, the component ranking attenuation weight is introduced so that the higher the ranking of the component in the formula, the higher the corresponding efficacy contribution weight. Several formulas with the highest similarity scores are extracted as candidate formulas. Common components in the candidate formulas are extracted, and a second comprehensive score is calculated for the common components using the ingredient efficacy database. An optimized ingredient ranking sequence is generated based on the comprehensive score in descending order. The corresponding benchmark proportion template is retrieved based on the ingredient ranking sequence and the target product dosage form to estimate the proportion of each ingredient and generate a recommended cosmetic formula.

[0113] In some embodiments, processor 501 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 502 or process data, such as executing a cosmetic formula recommendation program based on multi-dimensional features and dynamic calculation.

[0114] This embodiment also provides a computer-readable storage medium storing a cosmetic formula recommendation program based on multidimensional features and dynamic calculation. When the cosmetic formula recommendation program based on multidimensional features and dynamic calculation is executed by a processor, it implements the steps of the cosmetic formula recommendation method based on multidimensional features and dynamic calculation.

[0115] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0117] The technical solutions for collecting personal information involved in this embodiment strictly comply with the provisions of the "Personal Information Protection Law of the People's Republic of China" and Article 5 of the "Patent Law of the People's Republic of China," and do not violate any laws, social ethics, or harm public interests. The personal information collected by this technical solution does not originate from information collection devices in public places, and the explicit consent of each data collection subject has been obtained before information collection. The design and implementation of the technical solution do not involve batch image collection or identity recognition scenarios in public places, and do not require the application of special regulations for the installation of equipment in public places, but still strictly comply with the general national requirements for personal information protection, fully guaranteeing the data collection subjects' right to know, right to consent, and information security. The collected information is used only for the legal purposes stipulated in the solution, without any design for abuse or illegal transfer, and will not harm the legitimate rights and interests of the public or the normal social order, complying with the requirements of public order and good morals and the protection of public interests.

Claims

1. A method for recommending a cosmetic formulation based on multi-dimensional features and dynamic extrapolation, characterized by, include: Receive the user's structured requirements, transform the structured requirements into feature vectors, and construct a formula database for storing formula data and an ingredient efficacy database for storing ingredient efficacy quantification data. The feature vector is matched with the feature vector of the formula in the formula database to calculate the comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating the efficacy matching degree, the component ranking attenuation weight is introduced so that the higher the ranking of the component in the formula, the higher the corresponding efficacy contribution weight. Several formulas with the highest similarity scores are extracted as candidate formulas. Common components in the candidate formulas are extracted, and a second comprehensive score is calculated for the common components using the ingredient efficacy database. An optimized ingredient ranking sequence is generated based on the comprehensive score in descending order. The corresponding benchmark proportion template is retrieved based on the ingredient ranking sequence and the target product dosage form to estimate the proportion of each ingredient and generate a recommended cosmetic formula. 2.The method of claim 1, wherein, Also includes: The generated recommended cosmetic formulas are input into the regulatory engine for compliance verification. Once the verification is successful, a compliant recommended cosmetic formula is output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients. 3.The method of claim 1, wherein, The process of receiving the user's structured requirements and converting those requirements into feature vectors includes: The six-dimensional intelligent navigation input layer receives the user's structured needs, which includes a production plant layer, a site of action layer, a product dosage form layer, a usage method layer, an efficacy claim layer, and a user group layer. The structured requirements are transformed into feature vectors through a vectorization processing engine. The feature vectors include a basic framework vector, an efficacy weight vector, and a population vector. The basic framework vector is obtained by uniquely encoding the qualifications of the production plant, the site of action, and the product dosage form. The efficacy weight vector is extracted based on the main efficacy, secondary efficacy, and their weights. The population vector is extracted based on the characteristics of the user population. The formula database stores basic information, ingredient ranking sequence, efficacy labels, and target audience labels for historical compliant formulas and formulas currently on the market; the ingredient efficacy database stores quantitative data on ingredient efficacy, including the intended use of the ingredients, efficacy intensity matrix, and typical concentration range. 4.The method of claim 1, wherein, The feature vector is compared with the feature vectors in the recipe database using a multi-dimensional similarity matching calculation to obtain a comprehensive similarity score, including: Calculate the basic framework similarity, efficacy matching degree, and population matching degree between the feature vector and the formula feature vector; wherein, when calculating the efficacy matching degree, an ingredient ranking attenuation weight is introduced according to the ranking of each ingredient in the formula, so that the ingredient with the higher ranking has a higher efficacy contribution weight. The ingredient ranking attenuation weight is combined with the efficacy intensity data of the ingredient to calculate the individual efficacy matching degree. The overall efficacy matching degree is obtained by summing the individual efficacy matching degrees of all ingredients. Preset weights are assigned to the basic framework similarity, efficacy matching, and population matching. The similarity of each dimension is multiplied by the corresponding preset weight and then summed to obtain the comprehensive similarity score. 5.The method of claim 4, wherein, The overall similarity score is calculated using the following formula: Score total = w1 x Sim base + w2 x Sim eff + w3 x Sim aud In the formula, Score total is a comprehensive similarity score, Sim base is a base framework similarity, Sim eff is a function matching degree, Sim aud is a crowd matching degree, w1 is a base framework similarity weight, w2 is a function matching degree weight, and w3 is a crowd matching degree weight. Wherein the efficacy matching degree Sim eff A component rank attenuation weight is introduced in the calculation of the efficacy matching degree Sim, and the component rank attenuation weight is calculated by the following formula: In the formula, Weight rank is the component ranking decay weight, Rank is the ranking number of the component in the formula, the earlier the ranking, the smaller the Rank value, and the larger the Weight rank . The efficacy matching degree Sim eff The calculation method is as follows: for each ingredient in the cosmetic formula, the corresponding ingredient efficacy intensity of the ingredient in the ingredient efficacy database is multiplied by the ranking attenuation weight of the ingredient, to obtain the single-item efficacy score of the ingredient, and the single-item efficacy scores of all ingredients are summarized to obtain the efficacy matching degree Sim eff . 6.The method of claim 1, wherein, The process involves extracting several formulas with the highest similarity scores as candidate formulas, extracting common components from these candidate formulas, performing a secondary comprehensive score calculation on these common components using the component efficacy database, and generating an optimized component ranking sequence based on the comprehensive score in descending order. This includes: The comprehensive similarity scores are sorted from high to low, and the top-ranked formulas are extracted to form candidate formulas. The common components in the candidate formulas are then extracted. Combining preset efficacy weights, and using quantitative data of ingredients in the ingredient efficacy database, a secondary comprehensive score is calculated for the common ingredients. The secondary comprehensive score includes efficacy matching score, population suitability score, and dosage form suitability score. The common components are arranged in descending order of their secondary comprehensive scores to generate an optimized component ranking sequence. 7.The method according to claim 1 or 6, wherein, The step of retrieving the corresponding baseline proportion template based on the ingredient ranking sequence and the target product dosage form, calculating the proportion of each ingredient, and generating a recommended cosmetic formula includes: Based on the target product dosage form, retrieve the pre-stored benchmark proportion template corresponding to the dosage form, wherein the benchmark proportion template contains the benchmark proportion range corresponding to different ingredient rankings; Determine whether the ingredient efficacy database stores the typical concentration range of the corresponding ingredient; If the ingredient efficacy database stores the typical concentration range of the corresponding ingredient, and the typical concentration range falls within the benchmark proportion range corresponding to the ingredient's ranking, then the precise typical concentration value of the ingredient will be output. If the typical concentration range of the corresponding ingredient is not stored in the ingredient efficacy database, or if the typical concentration range exceeds the benchmark percentage range corresponding to the ranking of the ingredient, then the benchmark percentage range corresponding to the ranking of the ingredient will be output. By integrating the order and proportion of all ingredients, a recommended cosmetic formula is generated.

8. A multi-dimensional feature and dynamic inference based cosmetic formulation recommendation system for performing the multi-dimensional feature and dynamic inference based cosmetic formulation recommendation method of any one of claims 1-7, characterized in that, The system includes: The six-dimensional intelligent navigation input layer is used to receive users' six-dimensional structured requirements. The six-dimensional intelligent navigation input layer includes a production plant layer, a site of action layer, a product dosage form layer, a usage method layer, an efficacy claim layer, and a user group layer. A vectorization processing layer is used to transform the six-dimensional structured requirements into feature vectors; The core recommendation engine layer has a built-in formula database and ingredient efficacy database. It is used to perform multi-dimensional similarity matching calculations between the feature vector and the formula feature vector in the formula database to obtain a comprehensive similarity score. The multi-dimensional similarity includes basic framework similarity, efficacy matching degree and population matching degree. When calculating efficacy matching degree, an ingredient ranking attenuation weight is introduced so that the higher the ranking of an ingredient in the formula, the higher the corresponding efficacy contribution weight. The formula generation and estimation layer is used to extract several formulas with the highest similarity scores as candidate formulas, extract common components in the candidate formulas, perform a second comprehensive score calculation on the common components using the component efficacy database, generate an optimized component ranking sequence based on the comprehensive score in descending order, and retrieve the corresponding benchmark proportion template based on the component ranking sequence and the target product dosage form to estimate the proportion of each component and generate a recommended cosmetic formula. The output and compliance layer is used to input the generated cosmetic recommendation formulas into the regulatory engine for compliance verification. After the verification is passed, the compliant cosmetic recommendation formulas are output. The regulatory engine has built-in compliance verification rules, which include the highest historical usage limit of the catalog of used cosmetic ingredients and the safety standards for the use of cosmetic ingredients.

9. An electronic device, characterized in that Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the cosmetic formulation recommendation method based on multidimensional features and dynamic calculation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the cosmetic formulation recommendation method based on multidimensional features and dynamic calculation as described in any one of claims 1 to 7.

Images