Precise skin care personalized recommendation method and system based on artificial intelligence

Abstract

The invention discloses a precise skin care personalized recommendation method and system based on artificial intelligence. The method comprises the following steps: acquiring related data of a target user and data of a plurality of recommendable skin care products; preprocessing the attribute data to obtain user attribute features; preprocessing the user skin image to obtain user skin features; forming user cross features based on the user attribute features and the user skin features; preprocessing the recommendable skin care product data to obtain product features; inputting the user cross features and the product features into a product scoring model to obtain product scoring results corresponding to the plurality of skin care products; and determining a target recommended skin care product from the plurality of skin care products based on the product score result. The recommendation accuracy of the skin care products can be improved, the recommendation efficiency of the skin care products is improved, and then the recommendation effect of the skin care products is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a precise skin care personalized recommendation method and system implemented based on artificial intelligence. Background Art

[0002] To meet the intelligent needs of skin care products, it is necessary to actively recommend skin care products to users. The traditional method is to recommend skin care products based on fixed recommendation rules; however, the association rules between products are complex and changeable and difficult to fix, and the recommendation rules are difficult to enumerate, resulting in poor recommendation effects of products; at the same time, manually formulating recommendation rules leads to high labor costs and over-reliance on manual experience, resulting in poor recommendation effects of products.

[0003] Currently, most are directly determining the skin care products browsed by a user as the skin care products recommended to the user based on the user's historical browsing behavior of certain skin care products. However, since the frequency of user browsing is usually low, and even some skin care products may not be browsed again by the same user after being browsed once, and recommending such skin care products at this time will affect the user experience, that is, resulting in poor recommendation effects of skin care products. Summary of the Invention

[0004] The present invention provides a precise skin care personalized recommendation method and system implemented based on artificial intelligence to achieve highly accurate personalized precise recommendation of skin care.

[0005] A precise skin care personalized recommendation method implemented based on artificial intelligence includes: Obtaining relevant data of a target user and data of multiple recommendable skin care products, wherein the relevant data includes attribute data and a user skin image, and the user skin image is an unprocessed frontal facial image; Preprocessing the attribute data to obtain user attribute features; preprocessing the user skin image to obtain user skin features; Forming user cross features based on the user attribute features and the user skin features; Preprocessing the data of the recommendable skin care products to obtain product features; Inputting the user cross features and the product features into a product scoring model to obtain product score results corresponding to multiple skin care products, wherein the product scoring model is trained by a user feature set, a user cross feature set, a product feature set, and corresponding product score results; Determining a target recommended skin care product from the multiple skin care products based on the product score results.

[0006] As an implementable manner, the attribute data at least includes one or more of user basic information, user skin type information, and user behavior information; the user attribute characteristics at least include one or more of user basic characteristics, user skin type characteristics, and user behavior characteristics; Process the user basic information data to obtain user basic characteristics, where the user basic information includes gender and age; Process the user skin type information to obtain user skin type characteristics; Process the user behavior data to obtain user behavior characteristics, where the user behavior data includes one or more of the skin care products information browsed, browsing time data, activity data, transaction records, and user favorite data; the user behavior characteristics include one or more of skin care product characteristics, browsing time characteristics, activity characteristics, transaction record characteristics, and user favorite characteristics.

[0007] As an implementable manner, perform vectorization processing on the attribute data to obtain the user's attribute vector and then determine the user attribute characteristics; perform vectorization processing on the recommended skin care product data to obtain the attribute vector set of the skin care products and then determine the product characteristics.

[0008] As an implementable manner, the user cross characteristics are obtained through the following methods: Multiply the numerical fields in the user attribute characteristics and the user skin characteristics to generate new cross characteristics, or, Use one-hot encoding to transform and cross the user attribute characteristics and the user skin characteristics to generate new cross characteristics, or, Generate new cross characteristics by mapping the user attribute characteristics and the user skin characteristics to low-dimensional dense vectors and performing vector dot product, or, Create a polynomial combination of the user attribute characteristics and the user skin characteristics to generate new cross characteristics, or, Map the user attribute characteristics and the user skin characteristics to a fixed-size vector through feature hashing processing, and then create cross characteristics through the index of the hash bucket.

[0009] As an implementable manner, the user behavior data includes multiple sub-behavior data, determine the weights of each of the sub-behavior data, where the historical behavior data includes multiple sub-behavior data, and the sub-behavior data includes the number of times the user uses the product corresponding to the product characteristics, or, the cumulative number of purchases of the product corresponding to the product characteristics by the user, or, the cumulative number of searches for the product corresponding to the product characteristics by the user; based on the weights of each of the sub-behavior data, perform weighted aggregation processing on each of the sub-behavior data to obtain the user behavior data.

[0010] As an implementable mode, the product scoring model is trained through a user feature set, a user cross-feature set, a product feature set, and corresponding product score results, and includes the following steps: Construct a pre-trained product scoring model; Train the pre-trained product scoring model based on the user feature set, the user cross-feature set, the product feature set, and the corresponding product score results to obtain an initial product scoring model; Adjust the initial product scoring model through an objective function. When the result of the objective function reaches the preset requirement, the adjustment is completed to obtain the product scoring model; Among them, the product scoring model is expressed as follows:

[0011] Among them, represents the product scoring model, represents the intercept term of the product scoring model, represents the th user feature, represents the weight corresponding to the th user feature, represents the th product feature; represents the weight corresponding to the th product feature, , respectively represent the th user cross-feature and product feature, represents the weight corresponding to the th user cross-feature, represents transpose; The objective function is expressed as follows:

[0012] Among them, represents the loss function, representing the difference between the th true label and the product score result , represents the regularization term, represents the total number of trees, , the regularization term is expressed as follows:

[0013] Among them, represents the regularization result, represents the regularization coefficient of the tree structure complexity, represents the regularization coefficient of the leaf node weight, represents the number of leaf nodes of the tree, represents the weight of the th leaf node, represents the predicted product score, represents the total number of trees, i.e., the number of boosting rounds, the prediction function of the th tree, a decision tree model, , represents the bias term of the model, represents the number of leaf nodes of the th tree, represents the indicator function, represents a positive number, which takes the value of 1 when belongs to the th leaf node

[0014] As an implementable manner, it further includes the following steps: Recommend the target recommended skin care product to the target user; Obtain the feedback data of the target user on the target recommended skin care product; Based on the feedback data, determine the actual product score result of the target user on the target recommended skin care product; Based on the user attribute characteristics, the product feature data of the target recommended skin care product, and the actual product score result, update the product rating model.

[0015] A precise skin care personalized recommendation system implemented based on artificial intelligence, including: A data acquisition module, configured to acquire relevant data of a target user and data of multiple recommended skin care products, where the relevant data includes attribute data and a user skin image, and the user skin image is an unprocessed frontal facial image; A data processing module, configured to preprocess the attribute data to obtain user attribute characteristics; preprocess the user skin image to obtain user skin characteristics; form user cross characteristics based on the user attribute characteristics and the user skin characteristics; preprocess the data of the recommended skin care products to obtain product characteristics; A model training module, configured to input the user cross characteristics and the product characteristics into a product rating model to obtain product score results corresponding to multiple skin care products, where the product rating model is trained by a user feature set, a user cross feature set, a product feature set, and corresponding product score results; A product determination module, configured to determine a target recommended skin care product from the multiple skin care products based on the product score results.

[0016] An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements the precise skin care personalized recommendation method based on artificial intelligence as described above.

[0017] A non-transitory computer-readable storage medium, wherein a computer software program is stored in the storage medium, and when the computer software program is executed by a processor, it implements the precise skin care personalized recommendation method based on artificial intelligence as described above.

[0018] In the precise skin care personalized recommendation method based on artificial intelligence provided by the embodiments of the present invention, the product scoring model is trained through a user feature set, a user cross-feature set, a product feature set, and corresponding product score results, and can accurately obtain the product score results of target users for each skin care product, thereby improving the recommendation accuracy of skin care products, and further improving the recommendation effect of skin care products; at the same time, the user attribute features include first feature data determined based on the user skin image of the target user, that is, the user attribute features are determined based on the user skin image. Since the data involved in the user skin image is very accurate, the determination accuracy of the user attribute features can be improved, thereby improving the recommendation accuracy of skin care products, and finally improving the recommendation effect of skin care products. Moreover, the acquisition efficiency of the user skin image is high, so the determination efficiency of the user attribute features can be improved, and further the recommendation efficiency of skin care products can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic flowchart of the precise skin care personalized recommendation method based on artificial intelligence provided by the present invention; Figure 2 is a schematic structural diagram of the precise skin care personalized recommendation system based on artificial intelligence provided by the present invention; Figure 3 is a schematic diagram of an embodiment of the electronic device provided by the embodiments of the present invention; Figure 4 is a schematic diagram of an embodiment of the computer-readable storage medium provided by the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0022] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0023] The method in the embodiments of the present invention takes a recommendation system as the execution entity. Refer to Figure 1 , Figure 1 is a schematic flowchart of a precise skin care personalized recommendation method implemented based on artificial intelligence. A precise skin care personalized recommendation method implemented based on artificial intelligence includes: S100. Obtain relevant data of a target user and data of a plurality of recommendable skin care products. Among them, the relevant data includes attribute data and a user skin image, and the user skin image is an unprocessed frontal facial image; S200. Preprocess the attribute data to obtain user attribute features; preprocess the user skin image to obtain user skin features; S300. Form user cross features based on the user attribute features and the user skin features; S400. Preprocess the data of the recommendable skin care products to obtain product features; S500. Input the user cross features and the product features into a product scoring model to obtain product score results corresponding to a plurality of skin care products, where the product scoring model is trained by a user feature set, a user cross feature set, a product feature set, and the corresponding product score results; S600. Determine a target recommended skin care product from the plurality of skin care products based on the product score results.

[0024] It should be noted that the embodiments of the present invention may involve the use of user data. In actual applications, user-specific personal data can be used in the solutions described in the text within the scope permitted by applicable laws and regulations of the country where it is located (for example, with the explicit consent of the user, actual notification to the user, etc.).

[0025] Here, the target user is the user for whom skin care products are to be recommended. The multiple skin care products are skin care products that can be recommended to the user.

[0026] The multiple skin care products cover various types of users as much as possible, so that the finally determined preset product feature dataset includes the product feature data corresponding to any skin care product. In one embodiment, the skin care products may include, for example, facial cleanser, facial cleansing gel, lotion, toner, essence, emulsion, cream, sheet mask, sleeping mask, mud mask, sunscreen, eye cream, lip balm, lip scrub, etc. The multiple attributes of the skin care products may include the applicable population of the product, product category, product usage frequency, product type, product sub-type, etc.

[0027] The attribute data includes at least one or more of user basic information, user skin type information, and user behavior information; the user attribute features include at least one or more of user basic features, user skin type features, and user behavior features. The user basic information data is processed to obtain user basic features, where the user basic information includes gender and age; the user skin type information is processed to obtain user skin type features; the user behavior data is processed to obtain user behavior features, where the user behavior data includes one or more of the skin care product information browsed, browsing time data, activity data, transaction records, and user favorite data; the user behavior features include one or more of skin care product features, browsing time features, activity features, transaction record features, and user favorite features. That is to say, the user behavior data includes multiple sub-behavior data, and the weights of each of the sub-behavior data are determined, where the historical behavior data includes multiple sub-behavior data, and the sub-behavior data includes the number of times the user uses the product corresponding to the product feature, or the cumulative number of purchases of the product corresponding to the product feature by the user, or the cumulative number of searches for the product corresponding to the product feature by the user; based on the weights of each of the sub-behavior data, weighted aggregation processing is performed on each of the sub-behavior data to obtain user behavior data.

[0028] Specifically, in one embodiment, the multiple users cover various types of users as much as possible, so that the finally determined preset user attribute feature set includes the feature data corresponding to any user.

[0029] Here, the multiple attribute data of the user can include age, gender, skin type, acne mark condition, skin brightness, skin sensitivity, pore size, skin pigmentation degree, etc.

[0030] Considering that the attribute content of the user's attributes is not necessarily numerical data, it is necessary to convert the attribute content of all attributes into eigenvalue belonging to numerical type. If the original attribute content is numerical data, only the normalization process needs to be performed on this attribute content. For example, the eigenvalue of age is the actual age of the user; the eigenvalue of male gender is 0, and the eigenvalue of female gender is 1; the eigenvalue of dry skin type is 0, the eigenvalue of oily skin type is 1, and the eigenvalue of combination skin type is 2.

[0031] Further, after obtaining the user data of the user, that is, after obtaining the attribute content of the multiple attributes of the user, data preprocessing is performed on the attribute content of the multiple attributes. The data preprocessing may include but is not limited to: data cleaning, data interpolation processing, outlier processing, normalization, and feature extraction to provide a high-quality preset user attribute feature set. Among them, data cleaning specifically refers to removing duplicate data to ensure the uniqueness of the data. Data interpolation processing specifically refers to using interpolation methods (such as mean interpolation, median interpolation) or prediction models based on machine learning for filling. Outlier processing specifically refers to using methods such as binning to identify and correct or remove outliers. Normalization specifically refers to using normalization methods or normalization methods to convert data into a unified scale for continuous variables (such as age), and performing one-hot encoding or label encoding on categorical variables (such as gender, skin type, acne mark condition, skin brightness, skin sensitivity, pore size, skin pigmentation degree) to convert them into numerical data.

[0032] In one embodiment, the user attribute features are obtained through the following method: Perform vectorization processing on the attribute data to obtain the attribute vector of the user, and then determine the user attribute features.

[0033] The user attribute features are used to characterize the attribute content of at least one attribute of the target user. Any product score result is determined based on the user feature set, the user cross-feature set, and the product feature set, so that the obtained product score result is the scoring result of the target user for the skin care product characterized by the product feature data.

[0034] Based on the product score result, determine the target recommended skin care product from the multiple skin care products. Here, the user features, user cross-features, and product features are input into the product scoring model to obtain the product score result of the target user for the skin care product output by the product scoring model, that is, the product feature data of the multiple skin care products are respectively input into the product scoring model.

[0035] In one embodiment, the data of recommendable skin care products is vectorized to obtain an attribute vector set of skin care products, and then product features are determined. The user cross-features are obtained in the following manner: Multiply the numerical fields in the user attribute features and the user skin features to generate new cross-features, or, use one-hot encoding to transform and cross the user attribute features and the user skin features to generate new cross-features, or, map the user attribute features and the user skin features to low-dimensional dense vectors and perform dot product of vectors to generate new cross-features, or, create polynomial combinations of the user attribute features and the user skin features to generate new cross-features, or, map the user attribute features and the user skin features to fixed-size vectors through feature hashing, and then create cross-features through the indexes of hash buckets.

[0036] The introduction of cross-features can improve the performance of the model, and also enhance the generalization ability and interpretability of the model. In addition, cross-features can help the model capture the non-linear relationships between input features; by creating new cross-features, the feature space of the model is increased, which may improve the prediction accuracy of the model. In some cases, the model may have high bias due to insufficient feature space, and cross-features can reduce this bias, making the model closer to the real data distribution. Through cross-features, the model can learn more complex data patterns, which helps the generalization ability of the model on new data. Cross-features can help the model identify which feature combinations are most important for the prediction result, so as to perform effective feature selection, reduce the complexity of the model and the risk of overfitting. In addition, there may be high-dimensional data sets in the user attribute features and the user skin features, and cross-features can identify and utilize complex relationships that cannot be captured by individual features. Cross-features can also make the model better adapt to changes in data distribution, especially in the case of uneven or skewed data distribution.

[0037] In another embodiment, the user feature set, the user cross-feature set, and the product feature set are all input into the product scoring model to obtain the product score results of the target user for each skin care product, that is, the product feature data of multiple skin care products are input into the product scoring model together. Here, the user skin features are emphasized. The user cross-feature set is determined based on the user skin image of the target user. Since the functions of various current photo-taking software are becoming more and more diverse, the user skin image here is an untreated frontal facial image. In this way, more accurate user skin features can be obtained. For example, through the frontal facial image, skin type, acne mark situation, skin brightness, skin sensitivity, pore size, skin pigmentation degree, etc. can be obtained. Furthermore, the accuracy and efficiency of determining user attribute features are improved. User skin features can improve the accuracy of determining sample user attribute features, thereby improving the training effect of the product scoring model and ultimately improving the recommendation accuracy of skin care products; and improve the efficiency of determining sample user attribute features, thereby improving the training efficiency of the product scoring model.

[0038] Specifically, the product scoring model is trained through the user feature set, the user cross-feature set, the product feature set, and the corresponding product score results, and includes the following steps: Construct a product scoring pre-training model; Based on the user feature set, the user cross-feature set, the product feature set, and the corresponding product score results, train the product scoring pre-training model to obtain an initial product scoring model; Adjust the initial product scoring model through an objective function. When the result of the objective function reaches the preset requirement, the adjustment is completed to obtain the product scoring model; Among them, the product scoring model is expressed as follows:

[0039] Among them, represents the product scoring model, represents the intercept term of the product scoring model, represents the th user feature, represents the weight corresponding to the th user feature, represents the th product feature; represents the weight corresponding to the th product feature, , respectively represent the th user cross-feature and product feature, represents the weight corresponding to the th user cross-feature, represents the transpose; The objective function is expressed as follows:

[0040] in, represents the loss function, represents the True labels and product score results The difference between represents the regularization term, represents the total number of trees, , the regularization term is expressed as follows:

[0041] in, represents the regularization result, The regularization coefficient representing the structural complexity of the tree, represents the regularization coefficient of the leaf node weight, Represents the number of leaf nodes in the tree, Indicates The weight of a leaf node, represents the predicted product score, represents the total number of trees, i.e. the number of boosting rounds, No. The prediction function of a tree, a decision tree model, represents the bias term of the model, , Indicates The number of leaf nodes in a tree, represents the indicator function, represents a positive number, when Belong to Leaf nodes , the value is 1, otherwise it is 0.

[0042] Moreover, in another embodiment, the mean square error function can be used as the loss function of the product rating model, that is, the model is trained by minimizing the error between the actual product score result and the predicted product score result. By training the product rating model in the above manner, the product rating model can learn and mine the potential association between users and skin care products, so that the product rating model can determine the product score result of the user for any skin care product, that is, the corresponding product score result can be obtained for any target user or any skin care product.

[0043] In other embodiments, there are also the following steps: Determine target recommended skin care products from multiple skin care products based on the score results of each product. For example: Based on the sorting results of the product scores of each product from largest to smallest, determine a preset number of target recommended skin care products with the highest product scores from multiple skin care products. Or, based on the sorting results of the product scores of each product from largest to smallest, determine a preset proportion of target recommended skin care products with the highest product scores from multiple skin care products. For example, if the number of multiple skin care products is 10 and the preset proportion is 0.2, then there are 2 target recommended skin care products. Or, determine target recommended skin care products whose product scores are greater than a preset product score threshold from multiple skin care products based on the score results of each product. If there are multiple target recommended skin care products, based on the sorting results of the product scores of each product from largest to smallest, determine the recommendation order (recommendation priority) of the multiple skin care products to be recommended, so as to recommend the multiple skin care products to be recommended to the target user based on the recommendation order of the multiple skin care products to be recommended.

[0044] In the embodiments of the present invention, the target user to be recommended and multiple skin care products that can be recommended are determined. The user attribute characteristics of the target user and the product characteristic data of the multiple skin care products are input into the product scoring model to obtain the product score results of the target user for each skin care product output by the product scoring model, so as to determine target recommended skin care products from multiple skin care products based on the score results of each product. And the product scoring model is trained based on the sample user attribute characteristics and product characteristics, as well as the product score result labels corresponding to the sample user attribute characteristics and product characteristics. Thus, even if the target user has not browsed skin care products historically, or the target user has not browsed all skin care products historically, or the target user has browsed some skin care products few times, through the above method, the product scoring model can accurately obtain the product score results of the target user for each skin care product based on the user attribute characteristics of the target user and the product characteristic data of the multiple skin care products, thereby improving the recommendation accuracy of skin care products, and further improving the recommendation effect of skin care products; at the same time, the user attribute characteristics include the first characteristic data determined based on the user skin image of the target user, that is, the user attribute characteristics are determined based on the user skin image. Since the data involved in the user skin image is very accurate, the determination accuracy of the user attribute characteristics can be improved, and further the recommendation accuracy of skin care products can be improved, and finally the recommendation effect of skin care products can be improved. And the acquisition efficiency of the user skin image is high, so the determination efficiency of the user attribute characteristics can be improved, and further the recommendation efficiency of skin care products can be improved.

[0045] Based on any of the above embodiments, considering that if the user attribute features need to be determined based on the user skin image of the target user every time when recommending skin care products to the target user, it will lead to a decrease in the determination efficiency of the user attribute features, thereby affecting the recommendation efficiency of the skin care products, and ultimately affecting the recommendation effect of the skin care products; based on this, in this method, the user attribute features are determined based on the following method: determining the user attribute features corresponding to the target user from the preset user attribute feature set. Wherein, the preset user attribute feature set includes the feature data corresponding to multiple users, and the preset user attribute feature set is set in advance, so that the user attribute features corresponding to the target user can be directly determined from the preset user attribute feature set, thereby improving the determination efficiency of the user attribute features and ultimately improving the recommendation efficiency of the skin care products.

[0046] Wherein, the preset user attribute feature set is determined based on the following method, including: obtaining the attribute feature value sets of multiple users; the attribute feature value set of any user includes the feature values of multiple attributes of the user, and the feature value of any attribute is determined based on the attribute content of the attribute; the attribute feature set of any user includes the feature values determined based on the user skin image of the user. Vectorize each feature value in each attribute feature value set respectively to obtain the attribute vector sets of multiple users; the attribute vector set of any user includes the feature vectors of multiple attributes of the user. It should be understood that after the feature value of an attribute is vectorized, the feature vector of this attribute is obtained. Specifically, based on the attribute vector sets of multiple users, calculate the similarity between users through a similarity algorithm, and the similarity between a user and similar users should be greater than the preset similarity threshold.

[0047] Wherein, the similarity between users can be calculated using the following similarity calculation formula: ; Wherein, represents the similarity between user A and user B, represents the number of multiple attributes of the user, represents the th attribute feature vector of user A, represents the th attribute feature vector of user B, and should be the feature vectors of the same attribute.

[0048] Finally, based on the attribute vector sets of multiple users, determine the preset user attribute feature set; the feature data corresponding to any user included in the preset user attribute feature set is the attribute vector set.

[0049] In one embodiment, the preset user attribute feature set determined based on the attribute vector sets of multiple users also includes the attribute vector sets of multiple users.

[0050] In another embodiment, the duplicate removal process is performed on the attribute vector sets of multiple users to obtain a preset user attribute feature set including the attribute vector sets of N users. N is less than or equal to the number of multiple users. If N is less than the number of multiple users, there will be a situation where one attribute vector set in the preset user attribute feature set corresponds to multiple users; that is, one attribute vector set in the preset user attribute feature set matches one or more users. Based on this, the occupied space of the preset user attribute feature set can be reduced.

[0051] In another embodiment, the clustering process is performed on the attribute vector sets of multiple users to obtain M clustering clusters. M is less than or equal to the number of multiple users. Any clustering cluster includes at least one attribute vector set; for each clustering cluster, any attribute vector set in the clustering cluster is randomly selected as the attribute vector set of the clustering cluster, or the cluster center of the clustering cluster is used as the attribute vector set of the clustering cluster; the preset user attribute feature set is determined based on the attribute vector sets of each clustering cluster, that is, the preset user attribute feature set includes M attribute vector sets; that is, one attribute vector set in the preset user attribute feature set matches one or more users. Based on this, the occupied space of the preset user attribute feature set can be reduced.

[0052] In the embodiment of the present invention, the user attribute features corresponding to the target user are determined from the preset user attribute feature set, so that the user attribute features corresponding to the target user can be directly determined from the preset user attribute feature set, thereby improving the determination efficiency of the user attribute features and ultimately improving the recommendation efficiency of skin care products; and the preset user attribute feature set is determined in the above manner to ensure that the user attribute features are attribute vector sets, so that the input data input to the product scoring model is a feature vector, which is convenient for the product scoring model to better perform product scoring, and thus improves the recommendation accuracy of skin care products.

[0053] Based on any of the above embodiments, considering that if it is necessary to determine the product feature data every time a skin care product is recommended to a target user, the determination efficiency of the product feature data will be reduced, thereby affecting the recommendation efficiency of the skin care product, and ultimately affecting the recommendation effect of the skin care product; based on this, in this method, the product feature data of multiple skin care products is determined based on the following method: determining the product feature data corresponding to multiple skin care products from a preset product feature data set. Among them, the preset product feature data set includes the product feature data corresponding to multiple skin care products, and the preset product feature data set is set in advance, so that the product feature data corresponding to multiple skin care products can be directly determined from the preset product feature data set, thereby improving the determination efficiency of the product feature data and ultimately improving the recommendation efficiency of the skin care product. Among them, the preset product feature data set is determined based on the following method: obtaining the set of attribute feature values of multiple skin care products; the set of attribute feature values of any skin care product includes the feature values of multiple attributes of the skin care product, and the feature value of any attribute is determined based on the attribute content of the attribute; respectively performing vectorization processing on each feature value in each set of attribute feature values to obtain the set of attribute vectors of multiple skin care products; the set of attribute vectors of any skin care product includes the feature vectors of multiple attributes of the skin care product; based on the set of attribute vectors of multiple skin care products, determining the preset product feature data set; the product feature data corresponding to any skin care product included in the preset product feature data set is the set of attribute vectors.

[0054] Considering that the attribute content of the attributes of skin care products is not necessarily numerical data, it is necessary to convert the attribute content of all attributes into feature values belonging to numerical data. If the original attribute content is numerical data, it can be standardized directly.

[0055] In the embodiment of the present invention, the product feature data corresponding to multiple skin care products is determined from the preset product feature data set, so that the product feature data corresponding to multiple skin care products can be directly determined from the preset product feature data set, thereby improving the determination efficiency of the product feature data and ultimately improving the recommendation efficiency of the skin care product; and the preset product feature data set is determined in the above manner to ensure that the product feature data is a set of attribute vectors, so that the input data input to the product scoring model is a feature vector, which is convenient for the product scoring model to better perform product scoring, and ultimately improves the recommendation accuracy of the skin care product.

[0056] In one embodiment, the user feature vector and any product feature vector are input into the product scoring layer to obtain the product score result of the target user for the skin care product output by the product scoring layer, that is, multiple product feature vectors are respectively input into the product scoring layer.

[0057] In another embodiment, the user feature vector and multiple product feature vectors are both input into the product scoring layer to obtain the product score results of the target user for each skin care product, that is, multiple product feature vectors are input into the product scoring layer together.

[0058] In the embodiment of the present invention, the user attribute features are input into the low-dimensional feature extraction layer in the product scoring model to obtain the user feature vector output by the low-dimensional feature extraction layer. Each product feature data is respectively input into the low-dimensional feature extraction layer to obtain multiple product feature vectors output by the low-dimensional feature extraction layer. And the low-dimensional feature extraction layer is used to extract low-dimensional feature vectors, that is, to represent the compressed information of the original high-dimensional sparse vector in the low-dimensional space, so as to facilitate the subsequent product scoring layer to perform product scoring, and then improve the accuracy of product scoring, and finally improve the recommendation accuracy of skin care products.

[0059] In one embodiment, the historical behavior data includes multiple sub-behavior data, and the scoring results of each sub-behavior data are determined; the scoring results are aggregated to obtain the product score result label.

[0060] In the embodiment of the present invention, the historical behavior data of the sample user corresponding to the sample user attribute features is obtained to determine the product score result label corresponding to the sample user attribute features and product features based on the historical behavior data. And the historical behavior data is the behavior data of the sample user using the product corresponding to the product features, so that the product score result label can be accurately determined based on the actual historical behavior data of the sample user, improving the training effect of the product scoring model, and finally improving the recommendation accuracy of skin care products.

[0061] Based on the above embodiments, the historical behavior data includes multiple sub-behavior data. Based on the historical behavior data, determining the product score result label corresponding to the sample user attribute features and product features includes: determining the scoring results of each sub-behavior data; based on the weights of each sub-behavior data, performing weighted aggregation processing on each scoring result to obtain the user behavior data.

[0062] Exemplarily, the multiple sub-behavior data includes the number of times the sample user uses the product corresponding to the product features, the cumulative number of purchases of the product corresponding to the product features by the sample user, and the cumulative number of searches for the product corresponding to the product features by the sample user; if the usage time is less than 3 seconds, 1 point is obtained, if it exceeds 3 seconds, 2 points are obtained, and the weight of the usage time is 0.05; if the cumulative number of purchases is 0, 0 points are obtained, if it is 1, 1 point is obtained, and if it is 2 or more, 2 points are obtained, and the weight of the cumulative number of purchases is 0.4; if the cumulative number of searches is 0, 0 points are obtained, if it is 1, 2 points are obtained, and if it is 2 or more, 3 points are obtained, and the weight of the cumulative number of searches is 0.55. For example, if the scoring result of the usage time is 1 point, the score of the cumulative number of purchases is 1 point, and the score of the cumulative number of searches is 2 points, then the score corresponding to the product score result label is 1.55 points.

[0063] In the embodiment of the present invention, when the historical behavior data includes multiple sub-behavior data, it is considered that different sub-behavior data have different degrees of influence on the actual product score result. Based on this, based on the weights of each sub-behavior data, the scoring results are weighted and aggregated to obtain a more accurate product score result label, thereby improving the training effect of the product scoring model and ultimately improving the recommendation accuracy of skin care products.

[0064] Based on any of the above embodiments, in this method, the historical behavior data includes at least one of the following: the number of times a sample user uses the product corresponding to the product feature; the cumulative number of purchases of the product corresponding to the product feature by the sample user; the cumulative number of searches for the product corresponding to the product feature by the sample user. Here, the number of uses represents the cumulative number of times the user uses the product. The cumulative number of purchases represents the number of times the user purchases the product. The cumulative number of searches represents the number of times the user searches for the product. In the embodiment of the present invention, the historical behavior data includes the above three types of behavior data, so as to better mine the correlation between user behavior and products, obtain more accurate user behavior, thereby improving the training effect of the product scoring model, and ultimately improving the recommendation accuracy of skin care products.

[0065] Based on any of the above embodiments, the following steps are further included: Recommend the target recommended skin care product to the target user; Obtain the feedback data of the target user on the target recommended skin care product; Based on the feedback data, determine the actual product score result of the target user on the target recommended skin care product; Based on the user attribute characteristics, the product feature data of the target recommended skin care product, and the actual product score result, update the product scoring model.

[0066] It should be noted that after recommending skin care products to the target user, the target user is allowed to give feedback on the recommendation result, so as to collect the user's feedback data and continuously optimize the product scoring model, thereby improving the product scoring accuracy of the product scoring model and ultimately improving the recommendation accuracy of skin care products.

[0067] It should be noted that the feedback data can reflect the satisfaction of the target user with the recommended skin care product. Based on this, the actual product score result of the target user on the target recommended skin care product is determined based on the feedback data.

[0068] In one embodiment, the feedback data includes the scoring data of the target user on the recommended skin care product. Therefore, based on the scoring data, the actual product score result of the target user on the target recommended skin care product can be determined.

[0069] Based on the feedback data of the target user on the target recommended skin care product, the embodiment of the present invention determines the actual product score result of the target user on the target recommended skin care product, and continuously updates the product scoring model based on the user attribute characteristics, the product characteristic data of the target recommended skin care product and the actual product score result, so as to continuously improve the product scoring accuracy of the product scoring model, and further improve the recommendation accuracy of the skin care product, and finally improve the recommendation effect of the skin care product.

[0070] Refer to Figure 2 , Figure 2 which is a schematic structural diagram of a precise skin care personalized recommendation system implemented based on artificial intelligence provided by an embodiment of the present invention. The precise skin care personalized recommendation system implemented based on artificial intelligence includes: A data acquisition module 100, configured to acquire relevant data of the target user and data of multiple recommended skin care products. Among them, the relevant data includes attribute data and the user's skin image, and the user's skin image is an untreated frontal facial image; A data processing module 200, configured to preprocess the attribute data to obtain user attribute characteristics; preprocess the user's skin image to obtain user skin characteristics; form user cross characteristics based on the user attribute characteristics and the user skin characteristics; preprocess the data of the recommended skin care products to obtain product characteristics; A model training module 300, configured to input the user cross characteristics and the product characteristics into the product scoring model to obtain product score results corresponding to multiple skin care products. Among them, the product scoring model is trained by a user feature set, a user cross feature set, a product feature set and the corresponding product score results; A product determination module 400, configured to determine the target recommended skin care product from the multiple skin care products based on the product score results.

[0071] Please refer to Figure 3 , Figure 3 which is a schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. As Figure 3 shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, the following steps are implemented: S100. Acquire relevant data of the target user and data of multiple recommended skin care products. Among them, the relevant data includes attribute data and the user's skin image, and the user's skin image is an untreated frontal facial image; S200. Preprocess the attribute data to obtain user attribute characteristics; preprocess the user's skin image to obtain user skin characteristics; S300. Form user cross - features based on user attribute features and user skin features; S400. Pre - process the data of recommendable skin care products to obtain product features; S500. Input the user cross - features and product features into a product scoring model to obtain product score results corresponding to multiple skin care products, where the product scoring model is trained by a user feature set, a user cross - feature set, a product feature set, and the corresponding product score results; S600. Determine the target recommended skin care products from the multiple skin care products based on the product score results.

[0072] Please refer to Figure 4 , Figure 4 , which is a schematic diagram of an embodiment of a computer - readable storage medium provided by an embodiment of the present invention. As Figure 4 shown, this embodiment provides a computer - readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, the following steps are implemented: S100. Obtain relevant data of a target user and data of multiple recommendable skin care products, where the relevant data includes attribute data and a user skin image, and the user skin image is an untreated frontal facial image; S200. Pre - process the attribute data to obtain user attribute features; pre - process the user skin image to obtain user skin features; S300. Form user cross - features based on user attribute features and user skin features; S400. Pre - process the data of recommendable skin care products to obtain product features; S500. Input the user cross - features and product features into a product scoring model to obtain product score results corresponding to multiple skin care products, where the product scoring model is trained by a user feature set, a user cross - feature set, a product feature set, and the corresponding product score results; S600. Determine the target recommended skin care products from the multiple skin care products based on the product score results.

[0073] It should be noted that in the above - mentioned embodiments, the descriptions of each embodiment have their own focuses. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0074] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0075] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0076] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0078] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention.

[0079] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A precise skin care personalized recommendation method based on artificial intelligence, characterized in that: include: Obtain relevant data of the target user and multiple recommended skin care product data, wherein the relevant data includes attribute data and a user skin image, wherein the user skin image is an unprocessed frontal facial image; Preprocessing the attribute data to obtain user attribute features; preprocessing the user skin image to obtain user skin features; Forming user cross-features based on user attribute features and user skin features; Preprocess the recommended skin care product data to obtain product features; Inputting user cross-features and product features into a product scoring model to obtain product scoring results corresponding to a plurality of skin care products, wherein the product scoring model is trained by a user feature set, a user cross-feature set, a product feature set, and corresponding product scoring results; A target recommended skin care product is determined from the plurality of skin care products based on the product score result.

2. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 1, characterized in that: The attribute data includes at least one or more of user basic information, user skin quality information and user behavior information; the user attribute characteristics include at least one or more of user basic characteristics, user skin quality characteristics and user behavior characteristics; Processing the user's basic information data to obtain the user's basic characteristics, wherein the user's basic information includes gender and age; Processing the user's skin quality information to obtain the user's skin quality characteristics; The user behavior data is processed to obtain user behavior characteristics, wherein the user behavior data includes one or more of browsed skin care product information, browsing time data, activity data, transaction records and user collection data; the user behavior characteristics include one or more of skin care product characteristics, browsing time characteristics, activity characteristics, transaction record characteristics and user collection characteristics.

3. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 1, characterized in that: The attribute data is vectorized to obtain the user's attribute vector and then determine the user's attribute characteristics; the recommended skin care product data is vectorized to obtain the attribute vector set of the skin care product and then determine the product characteristics.

4. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 1, characterized in that: The user cross-feature is obtained in the following way: Multiply the numeric fields in the user attribute feature and the user skin feature to generate a new cross feature, or, Use one-hot encoding to transform and cross user attribute features and user skin features to generate new cross features, or, By mapping user attribute features and user skin features to low-dimensional dense vectors and performing vector dot multiplication to generate new cross features, or, Create a polynomial combination of user attribute features and user skin features to generate new cross features, or, User attribute features and user skin features are mapped to a fixed-size vector through feature hashing, and then cross-features are created through the index of the hash bucket.

5. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 2, characterized in that: The user behavior data includes multiple sub-behavior data, and the weight of each sub-behavior data is determined, wherein the historical behavior data includes multiple sub-behavior data, and the sub-behavior data includes the number of times the user uses the product corresponding to the product feature, or the cumulative number of times the user purchases the product corresponding to the product feature, or the cumulative number of times the user searches for the product corresponding to the product feature; based on the weight of each sub-behavior data, weighted aggregation processing is performed on each sub-behavior data to obtain the user behavior data.

6. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 1, characterized in that: The product rating model is obtained by training the user feature set, the user cross feature set, the product feature set and the corresponding product score results, and includes the following steps: Build a pre-trained model for product ratings; The product rating pre-training model is trained based on the user feature set, the user cross feature set, the product feature set and the corresponding product score results to obtain an initial product rating model; The initial product rating model is adjusted through the objective function. When the result of the objective function meets the preset requirements, the adjustment is completed and the product rating model is obtained. The product rating model is expressed as follows: in, represents the product rating model, represents the intercept term of the product rating model, Indicates User characteristics, Indicates The weight corresponding to each user feature, Indicates product features; Indicates The weights corresponding to the product features are , Respectively represent User cross-features and product features, Indicates The weight corresponding to the cross-feature of each user, represents transpose; The objective function is expressed as follows: in, represents the loss function, represents the True labels and product score results The difference between represents the regularization term, represents the total number of trees, , the regularization term is expressed as follows: in, represents the regularization result, The regularization coefficient representing the structural complexity of the tree, represents the regularization coefficient of the leaf node weight, Represents the number of leaf nodes in the tree, Indicates The weight of a leaf node, represents the predicted product score, represents the total number of trees, i.e. the number of boosting rounds, No. The prediction function of a tree, a decision tree model, represents the bias term of the model, , Indicates The number of leaf nodes in a tree, represents the indicator function, represents a positive number, when Belong to Leaf nodes , the value is 1, otherwise it is 0.

7. The method for personalized recommendation of precise skin care based on artificial intelligence according to claim 1, characterized in that: The following steps are also included: Recommending the target recommended skin care product to the target user; Obtaining feedback data from the target user on the target recommended skin care product; Based on the feedback data, determining an actual product score result of the target user for the target recommended skin care product; The product scoring model is updated based on the user attribute characteristics, the product feature data of the target recommended skin care product and the actual product score result.

8. An artificial intelligence-based personalized skin care recommendation system, characterized in that: include: A data acquisition module, used to acquire relevant data of a target user and data of a plurality of recommended skin care products, wherein the relevant data includes attribute data and a user skin image, wherein the user skin image is an unprocessed frontal facial image; A data processing module, used to pre-process the attribute data to obtain user attribute features; pre-process the user skin image to obtain user skin features; form user cross-features based on user attribute features and user skin features; pre-process the recommended skin care product data to obtain product features; A model training module, used for inputting user cross-features and product features into a product scoring model to obtain product scoring results corresponding to a plurality of skin care products, wherein the product scoring model is trained by a user feature set, a user cross-feature set, a product feature set and corresponding product scoring results; A product determination module determines a target recommended skin care product from the plurality of skin care products based on the product score result.

9. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the precise skin care personalized recommendation method based on artificial intelligence as described in any one of claims 1 to 7 when executing the program.

10. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, which, when executed by a processor, implements the precise personalized skin care recommendation method based on artificial intelligence as described in any one of claims 1 to 7.

Images