Determining Skin Complexion and Characteristics

1 . A method for predicting body complexion characteristics for body care comprising

predicting one or more objects in a plurality of first inputs at least by executing, at an extraction model of a trained feature learning model, an object extraction model of the trained feature learning model, wherein at least some of the plurality of first inputs are captured by an image capturing device of a mobile computing device and are transmitted as a first data stream to the extraction model;

receiving, at a classification model of the trained feature learning model, the one or more objects that are predicted by the extraction model and are transmitted to the classification model as a second data stream;

classifying, at a classification model of the trained feature learning model, the one or more objects into one or more respective classes, wherein a class of the one or more respective classes corresponds to a color index or value in a L*A*B* color space or a Lch color space;

predicting a body complexion index or value for the at least one subject area on the body of the client based at least in part upon the class, wherein the body complexion index or value represents a body complexion characteristic of the at least one subject area of the client; and

presenting, in a user interface of the mobile computing device, textual and graphical information pertaining to the body complexion index or value.

2 . The method of claim 1 further comprising:

identifying the plurality of first inputs, wherein the plurality of first inputs comprises one or more first actual images representing the at least one subject area on a body of a client.

3 . The method of claim 1 further comprising:

capturing the at least some of the plurality of first inputs during a scan session that invokes the image capturing device of a mobile computing device to generate the at least some of the plurality of first inputs.

4 . The method of claim 3 further comprising:

predicting, by an artificial intelligence model, a list of body care products or services based at least in part upon the body complexion index or value.

5 . The method of claim 3 wherein

the at least one subject area on the body of the client comprises a skin area on a skin of the client, a nail area on a nail of the client, or a hair area in the hairs of the client;

the body complexion index or value is predicted by the trained feature learning model and comprises a color tone or value for the at least one subject area,

the body complexion characteristic comprises a color, a size, a dimension, a shape, or a condition of the at least one subject area, and

the one or more objects comprise at least one of a hair, a freckle, a wrinkle, a mole, a differently colored spot, a pre-malignant body tissue growth, a malignant body tissue growth, or a capillary.

6 . The method of claim 1 further comprising:

identifying a plurality of second inputs, wherein the plurality of second inputs comprises one or more actual second images respectively representing subject areas on one or more bodies of one or more clients and one or more synthetic images; and

partition the plurality of second inputs into multiple subsets, the multiple subsets comprising a training subset that comprises the one or more actual images and the one or more synthetic images.

7 . The method of claim 6 further comprising:

identifying at least one actual image from the one or more actual images; and

transforming the at least one actual image into a synthetic image of the one or more synthetic images at least by applying a visibly imperceptible change to the at least one actual image.

8 . The method of claim 6 further comprising:

training, at a separate computing system, a feature learning model into the trained feature learning model using at least the training subset that is transmitted as a training data stream to the feature learning model and comprises the one or more actual images and the one or more synthetic images.

9 . The method of claim 8 wherein training the feature leaning model comprises:

training a plurality of model parameters of the feature learning model into a plurality of trained model parameters;

training a plurality of hyperparameters into a plurality of trained hyperparameters for the feature learning model; and

populating the plurality of trained parameters and the plurality of trained hyperparameters for the feature learning model.

10 . The method of claim 1 further comprising:

testing the trained feature learning model using at least a testing subset of the multiple subsets that is transmitted as a testing data stream to the trained feature learning model based at least in part upon a testing objective.

11 . A method for predicting products and recommendations for body care comprising

identifying a plurality of objects that comprises a plurality of user objects, a plurality of product or service objects, and a plurality of relationship objects, wherein the plurality of user objects includes a client object of a client, and the plurality of products or services includes multiple body care products or services;

determining a plurality of user latent representations for the plurality of user objects;

determining textual embeddings, visual embeddings, and relationship embeddings for the plurality of objects;

determining a plurality of product or service latent representations based at least in part upon the textual embeddings, the visual embeddings, and the relationship embeddings; and

predicting, by a joint learning model, a personalized, ranked recommendation for the client based at least in part upon ranking measures pertaining to at least the plurality of user latent representations and the plurality of product or service latent representations, wherein the personalized, ranked recommendation comprises information pertaining to one or more recommended body care products or services from the plurality of products or service for the client.

12 . The method of claim 11 wherein determining the textual embeddings, visual embeddings, and relationship embeddings comprises:

determining, by a relationship embedding model, a plurality of relationship embedding representations for the plurality of relationship objects.

13 . The method of claim 12 wherein determining the textual embeddings, visual embeddings, and relationship embeddings comprises:

determining, by a textual embedding model, a plurality of textual embedding representations for the plurality of user objects and the plurality of product and service objects; and

determining, by a visual embedding model, a plurality of visual embedding representations for the plurality of user objects and the plurality of product and service objects.

14 . The method of claim 12 wherein determining the plurality of relationship embedding representations further comprises:

determining one or more types of edges that interconnect the plurality of objects;

embedding the plurality of user objects and the plurality of product or service objects into respective embedding representations in an object space; and

embedding the plurality of relationship objects into respective relationship embedding representations in a relationship-specific object space.

15 . The method of claim 14 wherein determining the plurality of relationship embedding representations further comprises:

determining a scoring function and one or more constraints; and

determining a transform that maps the plurality of user objects and the plurality of product or service objects from the object space into the relationship-specific object space.

16 . The method of claim 15 wherein determining the plurality of relationship embedding representations further comprises:

identifying a user object from the plurality of user objects, a product or service object from the plurality of product or service objects, and a relationship object from the plurality of relationship objects into a triple, wherein an interaction is present between the user object and the product or service object; and

determining a destructed or incorrect triple based at least in part upon the triple.

17 . The method of claim 16 , wherein determining the destructed or incorrect triple comprises:

determining a plurality of pairs each comprising a respective user object of the plurality of user objects and a respective product or service object of the plurality of product or service objects; and

determining an interaction data structure including interaction data that indicates whether an interaction exists between the respective product or service object and the respective user object.

18 . The method of claim 16 wherein determining the destructed or incorrect triple further comprises:

replacing the user object with a different user object where no relationships or interactions exist between the different user object and the product or service object in the destructed or incomplete triple; or

replacing the product or service object with a different product or service object where no relationships or interactions exist between the user object and the different product or service object in the destructed or incomplete triple.

19 . The method of claim 18 wherein determining the plurality of relationship embedding representations further comprises:

determining a ranking measure for the triple and the destructed or incorrect triple using at least the scoring function and the one or more constraints based at least in part upon the interaction data structure; and

determining a Bayesian form for the triple and the destructed or incorrect triple using the scoring function.

20 . The method of claim 18 wherein determining the plurality of relationship embedding representations further comprises:

training the relationship embedding model using at least a plurality of triples and a plurality of corresponding destructed or incorrect triples with an objective function and a gradient descent algorithm.