A skin care efficacy vector quantization method and system fusing multi-dimensional constraints and semantic intent

Abstract

This invention relates to the field of natural language processing, and in particular provides a method and system for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent. The method includes receiving four-dimensional heterogeneous semantic input from users; constructing a multidimensional knowledge base, which includes a semantic weight association library, an intent-vector mapping library, and a constraint-correction matrix library; calculating a dynamic priority scalar for each skin problem by combining the influence of skincare motivation and skincare goals on various major skin problems; calling the intent-vector mapping library to obtain the primitive efficacy vectors corresponding to each skin problem, weighting the primitive efficacy vectors using the dynamic priority scalar, and aggregating them using a max-pooling algorithm to generate a baseline demand vector; calling the constraint-correction matrix library to obtain the corresponding correction matrix according to the constraints, performing nonlinear correction processing on the baseline demand vector, and outputting the final target efficacy representation vector. This invention achieves accurate, safe, and personalized quantification of skincare efficacy.

Description

Technical Field

[0001] This invention relates to the field of natural language processing, and in particular to a method and system for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent. Background Technology

[0002] With the widespread application of personalized recommendation technology in the beauty and skincare industry, how to accurately transform users' fuzzy and heterogeneous natural language needs into mathematical inputs that computers can process has become a key technical bottleneck for improving recommendation accuracy. Existing recommendation algorithms (especially advanced recommendation models based on vector algebra) heavily rely on high-quality numerical vectors as input.

[0003] Prior art 1, Chinese Patent Application No. CN202211000586.2, discloses a method for training and using a multi-intent semantic understanding model, an electronic device, and a storage medium. The method includes: encoding an input sentence labeled with joint tags to obtain a vector representation of the input sentence; parsing the vector representation to obtain an output sequence of joint tags; and training the multi-intent semantic understanding model based on the joint tag annotations and the output sequence of joint tags. By encoding an input sentence labeled with joint tags to obtain a vector representation of the input sentence, parsing the vector representation to obtain an output sequence of joint tags, and finally training the multi-intent semantic understanding model based on the joint tag annotations and the output sequence of joint tags, it is possible to represent and distinguish multiple intents with the same name appearing in the same sentence using joint tags. Prior art two, Chinese patent application number: CN202310812916.6, discloses a training method for a multi-domain, multi-intent spoken semantic understanding model. This method includes: inputting labeled ontology item data into the spoken semantic understanding model; semantically encoding the dialogue text based on the ontology encoding module within the spoken semantic understanding model to obtain initial ontology features and initial vectors; encoding the initial ontology features and initial vectors using a graph encoder based on a bidirectional relational graph attention network to obtain semantic and structural encodings; and inputting the semantic and structural encodings into the decoder within the spoken semantic understanding model to obtain the probability distribution of the spoken semantic understanding results. This invention, starting from the multi-intent spoken semantic understanding task, extends the spoken semantic understanding task to multi-domain settings and hierarchical semantic framework output. It constructs a multi-intent dataset derived from real-world industrial scenarios and effectively solves the multi-domain, multi-intent spoken semantic understanding task. Prior art three, Chinese patent application number: CN202010693872.6, discloses an improved semantic intent recognition method and LSTM architecture system, including the following steps: acquiring training corpus; Chinese word segmentation step; removing stop words and punctuation marks step; word vector calculation step; generating a corresponding m*n word vector matrix based on the number of feature words n and the word vector dimension m of each feature word in the training corpus, and inputting the word vector matrix into a stacked LSTM architecture to learn and train the training corpus; the stacked LSTM architecture is composed of multiple stacked LSTM layers, the first LSTM layer learns and trains the word vector matrix to generate a first feature value matrix after learning and training, the first feature value matrix is ​​used as the input of the next LSTM layer, the last LSTM layer learns and trains the feature value matrix output by the previous LSTM layer, thereby outputting a second feature value matrix after learning and training; the second feature value matrix is ​​classified by an external softmax function; Current technologies 1, 2, and 3 lack dynamic weighting based on multi-dimensional context. Existing models typically perform a 1:1 label matching based solely on the user's selected "skin problem." However, a user's "skincare motivation" (e.g., for social confidence vs. for health) and "skincare goals" (e.g., wanting quick results vs. wanting to maintain stability) significantly alter the urgency of solving the same skin problem. Current technologies cannot dynamically adjust these weights through mathematical models. Hard constraints and soft needs are difficult to coordinate; current technologies struggle to simultaneously handle "soft needs" (e.g., wanting whiter skin) and "hard constraints" (e.g., sensitive skin). The common approach is simple filtering, lacking a non-linear correction mechanism, failing to maximize the expression of efficacy needs while ensuring safety. There is a lack of standardized vector interfaces; downstream advanced recommendation algorithms (vector algebra recommendation) require precise numerical vectors as input, while human language is ambiguous. Currently, there is a lack of a method to "compile" heterogeneous semantic inputs into unified mathematical vectors.

[0004] Therefore, this invention provides a method and system for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent. Summary of the Invention

[0005] To achieve the above objectives, the present invention adopts the following technical solution: One aspect of the present invention provides a method for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent, comprising the following steps: Receive four-dimensional heterogeneous semantic input from the user, which includes at least: main skin problems, skin care motivation, skin care goals and constraints; Construct a multidimensional knowledge base, which includes: a semantic weight association library, an intent-vector mapping library, and a constraint-correction matrix library; Based on the semantic weight association library, and combined with the influence of skincare motivation and skincare goals on each major skin problem, the dynamic priority scalar of each skin problem is calculated; The intent-vector mapping library is called to obtain the primitive efficacy vectors corresponding to each skin problem. The primitive efficacy vectors are weighted using the dynamic priority scalar and aggregated using the max pooling algorithm to generate a baseline demand vector. The constraint-correction matrix library is invoked to obtain the corresponding correction matrix according to the constraint conditions. The baseline demand vector is then subjected to nonlinear correction processing to output the final target efficacy representation vector.

[0006] In one alternative implementation, the calculation of a dynamic priority scalar for each skin problem... The following formula is used: ,in, The base weight for the i-th skin problem; and These are preset weighting coefficients for the user's selected skincare motivations and skincare goals; and These are the matching coefficients between the i-th skin problem retrieved from the semantic weight association library and the skincare motivation and skincare goal, respectively, with the matching coefficients ranging from [0,1]. It is used to characterize the urgency of solving the skin problem in a specific context.

[0007] In one optional implementation, the benchmark demand vector is generated by aggregating using a max-pooling algorithm. The specific process includes: calculating the weighted power vector for the i-th skin problem. The calculation formula is: ,in Let i be the primitive efficacy vector corresponding to the i-th skin problem; For each dimension d in the vector, the value with the largest value in that dimension among all weighted power vectors is selected as the output value of that dimension, expressed by the formula: Where A, B, and N represent the number of skin problems input, preserving the expression of the strongest efficacy requirement when multiple problems coexist.

[0008] In one optional implementation, the nonlinear correction processing of the baseline demand vector includes: Based on the sensitive muscle type and sensitivity level in the constraints, a corresponding correction matrix is ​​matched from the constraint-correction matrix library. ; Calculate the corrected target efficacy representation vector This is achieved by performing element-wise multiplication of the Hadamard product between the baseline demand vector and the correction matrix. The formula is as follows: In this matrix, the values ​​of each dimension are attenuation coefficients, ranging from [0,1], used to reduce or suppress the values ​​of the efficacy dimension that do not meet the safety constraints.

[0009] In one optional implementation, the method for constructing the semantic weight association library includes: Establish a mapping between skincare motivation and skin problems. For motivations related to social confidence, set a high matching coefficient for overt skin problems. For motivations related to health management, set a high matching coefficient for barrier damage problems. Establish a mapping between skincare goals and skin problems. For goals that produce quick results, set a high matching coefficient for acute inflammation or short-term improvement problems. For goals that maintain skin stability, set a high matching coefficient for deep regulation problems. The matching coefficients are dynamically initialized and updated using expert scoring or statistical learning methods based on labeled datasets. and .

[0010] In one optional implementation, the mechanism for constructing and updating the constraint-correction matrix library includes: Define a multidimensional sensitivity space, which includes at least a barrier sensitivity dimension and an inflammation sensitivity dimension, with each dimension corresponding to a different sensitivity level value; For each type of active ingredient or efficacy, a safety tolerance model is constructed at different sensitivity levels. A correction matrix is ​​generated based on the safety tolerance model, and the element values ​​in the matrix are the recommendation strength coefficients for the corresponding efficacy dimensions. When the user-input constraints contain sensitive information in multiple dimensions, multiple correction matrices are fused by superimposing or taking the minimum value to generate the final correction matrix.

[0011] In one optional implementation, a preliminary semantic cleaning and standardization step is also included: The input raw text is segmented and entity recognized to extract key entities related to skin problems, motivations, goals, and constraints. Based on a pre-built thesaurus, extracted keywords are mapped to standard terms in the knowledge base; If no user-input constraints are detected, the correction matrix will be set to an all-1 matrix by default, indicating that no safety attenuation will be performed.

[0012] Another aspect of the present invention provides a skincare efficacy vector quantification system that integrates multidimensional constraints and semantic intent, comprising: The input interaction module is used to receive unstructured text or structured options provided by the user and parse them into a four-dimensional input vector; The quantization processing module is used to load the multidimensional knowledge base and perform dynamic priority scalar calculation, intention vector weighted aggregation, and constraint matrix correction operations. The knowledge base management module is used to store and maintain the semantic weight association library, intent-vector mapping library, and constraint-correction matrix library, and supports incremental data updates; The vector output interface is configured to serialize the generated target efficacy representation vector into a standard format data packet and transmit it to downstream recommendation algorithms or product matching systems.

[0013] In another aspect, the present invention provides an electronic device comprising: At least one memory stores computer-executable instructions non-transiently; At least one processor, configured to run the computer-executable instructions, The computer-executable instructions are executed by the processor to implement the aforementioned method for quantizing skincare efficacy vectors that integrates multidimensional constraints and semantic intent.

[0014] In another aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by at least one processor, implement the aforementioned method for quantizing skincare efficacy vectors that integrates multidimensional constraints and semantic intent.

[0015] This invention receives four-dimensional heterogeneous semantic input from users, which includes at least: major skin problems, skincare motivations, skincare goals, and constraints. It constructs a multi-dimensional knowledge base, including a semantic weight association library, an intent-vector mapping library, and a constraint-correction matrix library. Based on the semantic weight association library, and considering the influence of skincare motivations and goals on each major skin problem, it calculates a dynamic priority scalar for each skin problem. It then calls the intent-vector mapping library to obtain the primitive efficacy vectors corresponding to each skin problem, weights these vectors using the dynamic priority scalar, and aggregates them using a max-pooling algorithm to generate a baseline demand vector. Finally, it calls the constraint-correction matrix library to obtain the corresponding correction matrix based on the constraints, performs non-linear correction processing on the baseline demand vector, and outputs the final target efficacy representation vector. This invention achieves precise, safe, and personalized quantification of skincare efficacy. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart of a skincare efficacy vector quantization method that integrates multidimensional constraints and semantic intent, provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart illustrating the construction and updating mechanism of the constraint-correction matrix library for a skincare efficacy vector quantification method that integrates multidimensional constraints and semantic intent, as provided in Embodiment 2 of the present invention. Figure 3 This is a flowchart of the semantic cleaning and standardization steps in the pre-processing of a skincare efficacy vector quantification method that integrates multidimensional constraints and semantic intent, as provided in Embodiment 2 of the present invention. Figure 4 This is a flowchart of a computer verification method for a skincare efficacy vector quantization method that integrates multidimensional constraints and semantic intent, as provided in Embodiment 2 of the present invention. Figure 5 This is a flowchart of the dynamic priority calculation process for a skincare efficacy vector quantization method that integrates multidimensional constraints and semantic intent, as provided in Embodiment 2 of the present invention. Figure 6 This is a schematic diagram of vector generation and correction for a skincare efficacy vector quantification method that integrates multidimensional constraints and semantic intent, as provided in Embodiment 2 of the present invention. Figure 7 This is a system logic architecture diagram of a skincare efficacy vector quantization system that integrates multidimensional constraints and semantic intent, as provided in Embodiment 3 of the present invention. Figure 8 This is a block diagram of the electronic device provided in Embodiment 4 of the present invention; Figure 9 This is a block diagram of a computer-readable storage medium provided in Embodiment 4 of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be configured as a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be configured as a direct electrical connection, such as physical contact and electrical conduction between two components, or as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be configured as an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be configured as an electrical connection between two components in a non-contact manner, such as capacitive coupling between two components to transmit electrical signals.

[0020] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation in which the components are schematically placed in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and are configured to change accordingly based on the orientation of the components in the accompanying drawings.

[0021] Example 1: like Figure 1 As shown, this embodiment of the invention provides a method for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent, comprising the following steps: Step S100: Receive the user's four-dimensional heterogeneous semantic input, which includes at least the main skin problems, skin care motivation, skin care goals, and constraints. Step S200: Construct a multidimensional knowledge base, which includes: a semantic weight association library, an intent-vector mapping library, and a constraint-correction matrix library; Step S300: Based on the semantic weight association library, and combined with the influence of skincare motivation and skincare goals on each major skin problem, calculate the dynamic priority scalar of each skin problem; Step S400: Call the intent-vector mapping library to obtain the primitive efficacy vectors corresponding to each skin problem, use the dynamic priority scalar to weight the primitive efficacy vectors, and use the max pooling algorithm to aggregate and generate the baseline demand vector. Step S500: Call the constraint-correction matrix library, obtain the corresponding correction matrix according to the constraint conditions, perform nonlinear correction processing on the baseline demand vector, and output the final target efficacy representation vector.

[0022] In the above embodiments, a multi-dimensional contextual semantic graph is constructed to obtain the user's original skincare need data, which includes skin problem tags, skincare motivation text, and skincare target text. Based on the multi-dimensional contextual semantic graph, feature extraction and vectorization mapping are performed on the original skincare need data to generate skin problem feature vectors, motivation weight coefficient vectors, and target timeliness correction vectors, respectively. A hard constraint correction model based on nonlinear penalty is constructed to define the recommendation space according to the physiological feature constraints input by the user, generating a safety boundary mask. Through a soft and hard constraint collaborative mechanism, the skin problem feature vector, motivation weight coefficient vector, target timeliness correction vector, and safety boundary mask are dynamically weighted and fused to generate a standardized skincare target efficacy representation vector. The skincare target efficacy representation vector is input to the downstream recommendation algorithm to output a matching skincare solution. Example

[0023] like Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 As shown, based on Example 1, the steps provided in this embodiment of the invention include calculating the dynamic priority scalar for each skin problem. The following formula is used: ,in, The base weight for the i-th skin problem; and These are preset weighting coefficients for the user's selected skincare motivations and skincare goals; and These are the matching coefficients between the i-th skin problem retrieved from the semantic weight association library and the skincare motivation and skincare goal, respectively, with the matching coefficients ranging from [0,1]. It is used to characterize the urgency of solving the skin problem in a specific context.

[0024] Specifically, the benchmark demand vector is generated by aggregating data using the max pooling algorithm. The specific process includes: calculating the weighted power vector for the i-th skin problem. The calculation formula is: ,in Let i be the primitive efficacy vector corresponding to the i-th skin problem; For each dimension d in the vector, the value with the largest value in that dimension among all weighted power vectors is selected as the output value of that dimension, expressed by the formula: Where A, B, and N represent the number of skin problems input, preserving the expression of the strongest efficacy requirement when multiple problems coexist.

[0025] Specifically, the nonlinear correction processing of the baseline demand vector includes: matching the corresponding correction matrix from the constraint-correction matrix library according to the sensitive muscle type and sensitivity level in the constraint conditions. ; Calculate the corrected target efficacy representation vector This is achieved by performing element-wise multiplication of the Hadamard product between the baseline demand vector and the correction matrix. The formula is as follows: , In the correction matrix, the values ​​of each dimension are attenuation coefficients, ranging from [0,1], used to reduce or suppress the values ​​of the efficacy dimension that do not meet the safety constraints.

[0026] Specifically, the method for constructing the semantic weight association library includes: Establish a mapping between skincare motivation and skin problems. For motivations related to social confidence, set a high matching coefficient for overt skin problems. For motivations related to health management, set a high matching coefficient for barrier damage problems. Establish a mapping between skincare goals and skin problems. For goals that produce quick results, set a high matching coefficient for acute inflammation or short-term improvement problems. For goals that maintain skin stability, set a high matching coefficient for deep regulation problems. The matching coefficients are dynamically initialized and updated using expert scoring or statistical learning methods based on labeled datasets. and .

[0027] Specifically, the mechanism for constructing and updating the constraint-correction matrix library includes: Define a multidimensional sensitivity space, which includes at least a barrier sensitivity dimension and an inflammation sensitivity dimension, with each dimension corresponding to a different sensitivity level value; For each type of active ingredient or efficacy, a safety tolerance model is constructed at different sensitivity levels. A correction matrix is ​​generated based on the safety tolerance model, and the element values ​​in the matrix are the recommendation strength coefficients for the corresponding efficacy dimensions. When the user-input constraints contain sensitive information in multiple dimensions, multiple correction matrices are fused by superimposing or taking the minimum value to generate the final correction matrix.

[0028] Specifically, this also includes preliminary semantic cleaning and standardization steps: The input raw text is segmented and entity recognized to extract key entities related to skin problems, motivations, goals, and constraints. Based on a pre-built thesaurus, extracted keywords are mapped to standard terms in the knowledge base; If no user-input constraints are detected, the correction matrix will be set to an all-1 matrix by default, indicating that no safety attenuation will be performed.

[0029] In the above embodiments, the feature extraction based on multi-dimensional context-based dynamic weighting specifically includes: Sentiment polarity analysis and intent recognition are performed on the text of skincare motivation, and a motivation weight coefficient vector Wm is calculated and generated. If the motivation of "social confidence" is identified, the weight coefficient of the corresponding skin problem is set to the first preset high threshold; if the motivation of "health stability" is identified, the weight coefficient of the corresponding skin problem is set to the second preset low threshold. Time sensitivity analysis is performed on the text of skincare target, and a target timeliness correction vector Tg is calculated and generated. A time decay function f(t) is established, which maps the target of "quick effect" to a high-slope time growth factor and the target of "long-term stability" to a low-slope time growth factor.

[0030] The construction of a hard constraint correction model based on nonlinear penalties specifically includes: collecting users' physiological characteristic data as hard constraints, establishing a nonlinear mapping matrix Mprohibit between physiological states such as sensitive skin, pregnant women, and patients and prohibited ingredients; calculating the safety score Sscore=σ(Mprohibit, Ccomponent), where Ccomponent is the feature vector of the component to be evaluated, and σ is the Sigmoid activation function; setting a safety threshold δ, when Sscore<δ, generating a safety boundary mask, forcibly setting the corresponding efficacy dimension value to zero or imposing a large negative penalty, so as to physically isolate unsafe recommendation paths in the numerical space.

[0031] The expression for the soft and hard constraint coordination mechanism is: Vreq=(Vproblem⊙Wm⊕Tg)⊗Mmask Wherein, Vreq is the final skincare target efficacy representation vector, Vproblem is the skin problem feature vector, ⊙ represents the Hadamard product, which represents the point-by-point modulation of the importance of the skin problem by the motivation; ⊕ represents vector superposition, which represents the bias correction of the target timeliness on the overall demand; Mmask is a diagonal matrix generated by the safety boundary mask, ⊗ represents matrix multiplication, which uses a hard constraint matrix to nonlinearly truncate the high-weight but unsafe efficacy demand.

[0032] The method for constructing a multidimensional contextual semantic graph includes: defining entity nodes within the skincare domain, which at least include: symptom nodes, ingredient nodes, mechanism nodes, motivation nodes, and scenario nodes; establishing semantic edge relationships between nodes, which include "alleviate", "cause", "applicable to", "conflict with", and "enhance"; initializing and embedding nodes using a pre-trained large-scale language model, and capturing cross-node dependencies in user demand texts through an attention mechanism to dynamically update the vector representations of entity nodes.

[0033] It also includes vector standardization and alignment steps: constructing an efficacy dimension space, which pre-defines N standardized efficacy axes, each axis corresponding to a quantified skincare efficacy index; projecting the generated skincare target efficacy representation vectors into the efficacy dimension space, and compiling the heterogeneous semantic inputs into fixed-dimensional real number vectors through a fully connected layer, where each dimension of the real number vector represents the intensity of the user's normalized demand for this type of efficacy.

[0034] The calculation of real-time numerical vectors also includes conflict resolution logic: when there are conflicting efficacy dimensions (such as "powerful exfoliation" and "barrier repair") in the skin care target efficacy representation vector and the values ​​of both are higher than the activation threshold, the conflict resolution module is activated. The conflict resolution module determines priorities based on the physiological characteristic data in the hard constraint correction model, reduces the values ​​of low-priority efficacy dimensions until the conflict evaluation value is below the preset range, thereby resolving the contradiction between soft needs and hard constraints.

[0035] Feature extraction employs a bidirectional encoder based on the Transformer architecture as the feature extractor. The feature extractor is fine-tuned using a domain-specific corpus containing skincare contextual data to capture the "user's implicit intent".

[0036] Furthermore, data input and graph mapping: the system parses the text and extracts key entities. Skin problems: [Acne: 0.9]; Motivation: [Social confidence: 0.95]; Target: [Quick results: 0.9]; Strict constraints: [Sensitive skin: Severe]; Multidimensional contextual dynamic weighting: Motivation weighting: Due to the detection of the "social confidence" motivation, the system calls the model parameters and increases the default urgency weight Wm of "acne" from the default 0.5 to 0.9 (emphasizing that this problem must be solved).

[0037] Target correction: If "immediate improvement" is detected, a time correction vector Tg is generated, and a positive bias is superimposed on the "significance of efficacy" dimension, which means that users are willing to take certain risks for high efficacy (but within a safe range).

[0038] Hard constraint correction: The system identifies "severely sensitive skin" and loads the security matrix Mprohibit.

[0039] Matrix analysis revealed that while users urgently need to get rid of acne, high concentrations of acidic ingredients (such as high concentrations of salicylic acid) pose a high risk to sensitive skin.

[0040] Generate a safe boundary mask Mmask that non-linearly truncates the numerical channels of the "high-irritation acne treatment" dimension (e.g., locks the upper limit at a low level) while fully opening the channels of the "soothing and anti-inflammatory" dimension.

[0041] Synergistic integration of soft and hard constraints: The core formula for calculation is: Vreq = (Vproblem⊙Wm⊕Tg)⊗Mmask; Before the integration, users' "acne removal" demand value may be as high as 0.98 (very strong).

[0042] After matrix multiplication with Mmask (sensitive skin constraint), the final "strong acne treatment" value was corrected to 0.4 (within the safe range), while the "soothing acne treatment" value was retained and increased to 0.8.

[0043] Results: The generated standardized efficacy vector Vreq is represented as: [Soothing: 0.9, Anti-inflammatory: 0.85, Strong exfoliation: 0.1, Moisturizing: 0.7…].

[0044] Recommended output: The downstream vector algebra recommendation algorithm calculates the cosine similarity between this vector and the product vectors in the product library. The system will not recommend "high-concentration salicylic acid exfoliating cotton" (although it treats acne quickly, the vector distance is far), but will recommend "centella asiatica soothing essence" (it highly matches the user-generated Vreq in terms of soothing and anti-inflammatory dimensions, meets the hard constraint of sensitive skin, and satisfies the basic skin improvement needs under social motivation). Example

[0045] like Figure 7 As shown, based on Example 1, this embodiment of the invention provides a skincare efficacy vector quantization system that integrates multidimensional constraints and semantic intent, including: The input interaction module is used to receive unstructured text or structured options provided by the user and parse them into a four-dimensional input vector; The quantization processing module is used to load the multidimensional knowledge base and perform dynamic priority scalar calculation, intention vector weighted aggregation, and constraint matrix correction operations. The knowledge base management module is used to store and maintain the semantic weight association library, intent-vector mapping library, and constraint-correction matrix library, and supports incremental data updates; The vector output interface is configured to serialize the generated target efficacy representation vector into a standard format data packet and transmit it to downstream recommendation algorithms or product matching systems.

[0046] In the above embodiments, this system is deployed on a cloud server and interacts with the user's front-end APP through an API interface. The database stores a skincare knowledge graph, an ingredient safety database, and a product vector library.

[0047] User skincare plan generation: Data Input: User B fills out the questionnaire and inputs the following information on the front end: Skin condition: "Dull complexion with blemishes", "Oily skin"; Psychological / Motivation: "I have an interview coming up soon and I want to look presentable." Expected goal: "To see significant changes within three days"; Physiological constraints: "Preparing for pregnancy".

[0048] Internal system processing flow: Work of the semantic intent parsing and vectorization module: The motivation analysis unit identified the keywords "interview" and "spiritual point," determining them as the **"social confidence / emergency social interaction"** motivation. The system then invoked the mapping table and set the weight coefficient Wm for the skin issue "whitening / brightening" to 0.95 (extremely high weight).

[0049] The target timeliness analysis unit identifies targets that are "within three days" and "significantly changed," classifying them as **"immediately effective" targets. The system generates an aggressive target timeliness correction vector Tg using a time decay function, which adds points to the "high concentration of active ingredients" dimension.

[0050] Hard constraint safety modeling module operation: The forbidden component mapping unit retrieved the physiological characteristic of "preparing for pregnancy".

[0051] Based on its internal medical safety matrix, the system identified common "whitening and brightening" drugs such as retinol (A-ol), high-concentration vitamin C and its derivatives as being included in the "high-risk / prohibited list" during pregnancy.

[0052] The mask generation unit generates a safe boundary mask Mmask. This mask is a diagonal matrix in which the dimensions corresponding to "retinol-based effects" and "acid-based skin rejuvenation effects" are set to a minimum value (e.g., 0.01) or zero, representing that the channel is closed; while the dimensions corresponding to "plant-based whitening" and "antioxidant" remain open.

[0053] Integration of quantitative calculation module (core innovation): If the synergy of soft and hard constraints is not considered, the value of the user's "whitening" demand vector may be as high as 0.98 (very strong).

[0054] Fusion computing: Vreq=(Vproblem⊙0.95⊕Tg)⊗Mmask During the calculation process, although the "whitening" demand value calculated in the first half is very large, the system will perform non-linear truncation when performing matrix multiplication with Mmask.

[0055] Specifically, although users want "quick whitening" (soft demand), because they are in the "preparation period" (hard constraint), the vector dimension representing the "high-risk whitening path" is strongly suppressed.

[0056] To satisfy users' social motivations, the system utilizes a "safe channel" opened within the mask to boost the values ​​of the "gentle brightening" and "antioxidant" dimensions as compensation. Furthermore, the system incorporates a conflict resolution and compensation mechanism: during non-linear correction processing, if a high-efficacy dimension (such as acid exfoliation) is detected to be truncated to zero or below a preset threshold due to safety constraints, the compensation module is automatically triggered. This module boosts the weight values ​​of the 'soothing / repairing' efficacy dimension related to the skin problem based on a semantic weight association library, maximizing the satisfaction of users' skincare motivations while ensuring safety.

[0057] Conflict resolution and output: The conflict resolution subsystem detected that the user wanted "quick results," but this logically conflicted with the safety restrictions during the "pre-conception period." Based on the preset rule: safety > efficacy, the system forcibly lowered the score for the "speed of results" dimension.

[0058] The vector normalization unit compiles the above calculation results into a standard vector output: Vreq=[Brightening: 0.88, Gentle: 0.95, Antioxidant: 0.85, Retinol: 0.00, Acid-based: 0.00, Moisturizing: 0.70] The vector is then sent to the downstream recommendation algorithm.

[0059] Downstream algorithms searched the product database and found that while the classic "high-concentration retinol night cream" had a good brightening effect, it was severely mismatched with the "retinol: 0.00" in the generated Vreq and was therefore filtered out. The algorithm ultimately calculated that this vector had the highest similarity to "a gentle whitening serum containing niacinamide and vitamin C derivatives" and recommended this product to user B. This recommendation responded to the user's strong motivation to "whiten skin for job interviews" while strictly adhering to the hard constraint of "safety during the pre-pregnancy period," achieving a balance between intelligence and safety.

[0060] Example 4: Figure 8 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0061] The electronic device is configured to include a central processing unit / microprocessor / main control chip, etc. 4; and a storage medium 5, coupled to the central processing unit / microprocessor / main control chip, etc. 4, and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0062] The central processing unit / microprocessor / main control chip, etc., are configured as including but not limited to one or more processors or microprocessors.

[0063] Storage medium 5 is configured to include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CDROM, DVDROM, Blu-ray disc, etc.).

[0064] In addition, the electronic device is also configured to include (but is not limited to) a data bus 6, an input / output bus / external bus / device bus 7, a display 8, and input / output devices 9 (e.g., keyboard, mouse, speaker, etc.).

[0065] The central processing unit / microprocessor / main control chip, etc., 4 are configured to communicate with external devices (8, 9, etc.) via I / O bus 7 through wired or wireless network (not shown).

[0066] The storage medium 5 is also configured to store at least one computer-executable instruction for performing steps of various functions and / or methods in the embodiments described herein when the central processing unit / microprocessor / main control chip, etc., 4 is running.

[0067] In one embodiment, the at least one computer-executable instruction is also configured to be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0068] Figure 9 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0069] like Figure 9As shown, the non-transitory computer-readable storage medium 11 stores instructions, such as computer-readable instructions 10. When the computer-readable instructions 10 are executed by a processor, they are configured to perform the various methods described above. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory is configured, for example, to include random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory is configured, for example, to include read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 11 is configured to be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 10 stored on the computer-readable storage medium 11, it is configured to perform the various methods described above.

[0070] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods are configured to be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, it may be configured in another way. For example, multiple units or components may be configured to be combined or used to integrate into another system, or some features may be configured to be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed is through some interface, or the indirect coupling or communication connection of the apparatus or unit may be configured to be electrical, mechanical, or other forms.

[0071] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, in the various embodiments of the present invention, the functional units are configured to be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it is configured to be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, is implemented as a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods of the various embodiments of the present invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media used for storing program code.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended 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 it is still used to modify the technical solutions described in the foregoing embodiments, or to make equivalent substitutions for some of the technical features. Such 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.

Claims

1. A method for quantifying skincare efficacy vectors by integrating multidimensional constraints and semantic intent, characterized in that, Includes the following steps: Receive four-dimensional heterogeneous semantic input from the user, which includes at least the main skin problems, skin care motivation, skin care goals and constraints. Based on the impact of the skincare motivations and goals on each major skin problem, a dynamic priority scalar for each skin problem is calculated. The primitive effectiveness vector is weighted using the dynamic priority scalar, and a benchmark demand vector is generated by aggregating the vectors using a max pooling algorithm. Based on the constraints, the corresponding correction matrix is ​​obtained, and the baseline demand vector is subjected to nonlinear correction processing to output the final target efficacy representation vector.

2. The skincare efficacy vector quantization method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, The dynamic priority scalar for calculating each skin problem The following formula is used: ,in, The base weight for the i-th skin problem; and These are preset weighting coefficients for the user's selected skincare motivations and skincare goals; and These are the matching coefficients between the i-th skin problem retrieved from the semantic weight association library and the skincare motivation and skincare goal, respectively, with the matching coefficients ranging from [0,1]. The coefficients are used to characterize the urgency of solving a skin problem in a specific context. They are dynamically initialized based on the experience scores of skincare experts or statistical learning of historical sample data.

3. The skincare efficacy vector quantization method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, The benchmark demand vector is generated by using the maximum pooling algorithm. The specific process includes: calculating the weighted power vector for the i-th skin problem. The calculation formula is: ,in Let i be the primitive efficacy vector corresponding to the i-th skin problem; For each dimension d in the vector, the value with the largest value in that dimension among all weighted power vectors is selected as the output value of that dimension, expressed by the formula: Where A, B, and N represent the number of skin problems input, this function aims to preserve the expression of the strongest efficacy requirement when multiple problems coexist, preventing the requirement from being diluted.

4. The skincare efficacy vector quantification method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, The nonlinear correction process for the baseline demand vector includes: Based on the sensitive muscle type and sensitivity level in the constraints, a corresponding correction matrix is ​​matched from the constraint-correction matrix library. ; Calculate the corrected target efficacy representation vector This is achieved by performing element-wise multiplication of the Hadamard product between the baseline demand vector and the correction matrix. The formula is as follows: In this matrix, the values ​​for each dimension are attenuation coefficients, ranging from [0,1], used to reduce or suppress the values ​​of efficacy dimensions that do not meet safety constraints. This achieves precise, targeted dimensionality reduction for specific efficacy dimensions, rather than changing the global weights.

5. The skincare efficacy vector quantization method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, The method for constructing the semantic weight association library includes: Establish a mapping between skincare motivation and skin problems. For motivations related to social confidence, set a high matching coefficient for overt skin problems. For motivations related to health management, set a high matching coefficient for barrier damage problems. Establish a mapping between skincare goals and skin problems. For goals that produce quick results, set a high matching coefficient for acute inflammation or short-term improvement problems. For goals that maintain skin stability, set a high matching coefficient for deep regulation problems. The matching coefficients are dynamically initialized and updated using expert scoring or statistical learning methods based on labeled datasets. and .

6. The skincare efficacy vector quantization method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, The mechanism for constructing and updating the constraint-correction matrix library includes: Define a multidimensional sensitivity space, which includes at least a barrier sensitivity dimension and an inflammation sensitivity dimension, with each dimension corresponding to a different sensitivity level value; For each type of active ingredient or efficacy, a safety tolerance model is constructed at different sensitivity levels. A correction matrix is ​​generated based on the safety tolerance model, and the element values ​​in the matrix are the recommendation strength coefficients for the corresponding efficacy dimensions. When the user-input constraints contain sensitive information in multiple dimensions, multiple correction matrices are fused by superimposing or taking the minimum value to generate the final correction matrix.

7. The skincare efficacy vector quantization method integrating multidimensional constraints and semantic intent as described in claim 1, characterized in that, It also includes preliminary semantic cleaning and standardization steps: The input raw text is segmented and entity recognized to extract key entities related to skin problems, motivations, goals, and constraints. Based on a pre-built thesaurus, extracted keywords are mapped to standard terms in the knowledge base; If no user-input constraints are detected, the correction matrix will be set to an all-1 matrix by default, indicating that no safety attenuation will be performed. If the constraint conditions in the four-dimensional heterogeneous semantic input are not detected, the correction matrix is ​​set to an all-1 matrix by default.

8. A skincare efficacy vector quantization system that integrates multidimensional constraints and semantic intent, as described in any one of claims 1 to 7, is characterized in that... include: The input interaction module is used to receive unstructured text or structured options provided by the user and parse them into a four-dimensional input vector; The quantization processing module is used to load the multidimensional knowledge base and perform dynamic priority scalar calculation, intention vector weighted aggregation, and constraint matrix correction operations. The knowledge base management module is used to store and maintain the semantic weight association library, intent-vector mapping library, and constraint-correction matrix library, and supports incremental data updates; The vector output interface is configured to serialize the generated target efficacy representation vector into a standard format data packet and transmit it to downstream recommendation algorithms or product matching systems.

9. An electronic device, comprising: At least one memory stores computer-executable instructions non-transiently; At least one processor, configured to run the computer-executable instructions, The computer-executable instructions are executed by the processor to implement the skincare efficacy vector quantization method according to any one of claims 1-7, which integrates multidimensional constraints and semantic intent.

10. A computer-readable storage medium, wherein, The computer-readable storage medium stores computer-executable instructions, which, when executed by at least one processor, implement a skincare efficacy vector quantization method that integrates multidimensional constraints and semantic intent according to any one of claims 1-7.

Images