# US 12,222,998 B2 · Methods and systems for determining a query response using a computing platform in a domain of beauty or cosmetic industry > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,222,998 B2 - Document ID: US12222998B2 - Status: Granted - Kind: B2 - Application number: 17/986,299 - Filed: 2022-11-14 - Published: 2024-05-16 - Granted: 2025-02-11 - Earliest claimed: 2022-11-14 - Assignees: Ulta Salon, Cosmetics & Fragrance, Inc. - Applicants: Ulta Salon, Cosmetics & Fragrance, Inc. - Inventors: Ben Pham; Chatura Samarasinghe; David Ruvalcaba - Web page: https://patentgrep.com/patent/US12222998B2 ## CPC - G06F 16/3329: Natural language query formulation - G06F 16/9532: Query formulation - G06F 16/9535: Search customisation based on user profiles and personalisation - G06F 16/9538: Presentation of query results - G06N 3/045: Combinations of networks - G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound - G06N 20/20: Ensemble learning - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract An example method includes receiving by a server and from a computing device a natural language-based query, converting the natural language-based query into an embedding formatted query having a vector format, inputting the embedding formatted query into a machine-learning logic model, and determining by the machine-learning logic model a query response to the embedding formatted query that is predicted by the machine-learning logic model to be in context of beauty or cosmetic industry. The machine-learning logic model is trained using a multi-dimensional learned-embedding that includes semantically similar terms in proximity in an embedding space and the embedding space is limited to salient terms associated with the beauty or cosmetic industry. The method also includes communicating by the machine-learning logic model the query response to the computing device. ## Claims ### Claim 1 A method of providing a computing platform in a domain of beauty or cosmetic industry, comprising: receiving, by a server and from a computing device, a natural language-based query; converting the natural language-based query into an embedding formatted query having a vector format; inputting the embedding formatted query into a machine-learning logic model; determining, by the machine-learning logic model, a query response to the embedding formatted query that is predicted by the machine-learning logic model to be in context of beauty or cosmetic industry, wherein the machine-learning logic model is trained using a multi-dimensional learned-embedding that includes semantically similar terms in proximity in an embedding space, wherein the embedding space is limited to salient terms associated with the beauty or cosmetic industry, wherein determining, by the machine-learning logic model, the query response to the embedding formatted query comprises: determining, by a first machine-learning logic model, a first predicted intent to the embedding formatted query, wherein the first predicted intent is predicted by the first machine-learning logic model to be in context of beauty or cosmetic industry, wherein the first machine-learning logic model is trained using data limited to salient terms associated with the beauty or cosmetic industry; determining, by a second machine-learning logic model, a second predicted intent to the embedding formatted query, wherein the second predicted intent is different than the first predicted intent, wherein the second predicted intent is predicted using the second machine-learning logic model that is trained using data not limited to salient terms associated with the beauty or cosmetic industry; combining the first predicted intent and the second predicted intent into a feature vector; and processing, by the machine-learning logic model, the feature vector to determine the query response; and communicating, by the machine-learning logic model, the query response to the computing device. ### Claim 2 (depends on 1) The method of claim 1, wherein communicating, by the machine-learning logic model, the query response to the computing device comprises: based on receiving the natural language-based query from a customer service platform, communicating a determined intent action item and an associated subject to the computing device. ### Claim 3 (depends on 1) The method of claim 1, wherein communicating, by the machine-learning logic model, the query response to the computing device comprises: based on receiving the natural language-based query from a search platform, communicating a determined item of interest to a search engine; and outputting search results from the search engine to the computing device. ### Claim 4 (depends on 1) The method of claim 1, wherein converting the natural language-based query into the embedding formatted query having the vector format comprises: mapping words of the natural language-based query into the vector format comprising a series of numbers such that similar meaning words are located together and dissimilar words are located further apart. ### Claim 5 (depends on 1) The method of claim 1, further comprising: pre-processing the natural language-based query to remove non-descriptive words from the natural language-based query prior to converting the natural language-based query into the embedding formatted query. ### Claim 6 (depends on 1) The method of claim 1, wherein processing, by the machine-learning logic model, the feature vector to determine the query response comprises: processing the feature vector using a gradient-boosted decision tree machine learning library. ### Claim 7 (depends on 1) The method of claim 1, wherein the first machine-learning logic model includes a first level model and a second level model, and the method further comprises: training the first level model in the first machine-learning logic model using first level categories; and training the second level model in the first machine-learning logic model using second level categories more narrow than the first level categories. ### Claim 8 (depends on 7) The method of claim 7, wherein each of the first level model and the second level model are divided into multiple models, and each of the multiple models are trained differently based on a size of a training data set. ### Claim 9 (depends on 1) The method of claim 1, wherein: determining, by the first machine-learning logic model, the first predicted intent to the embedding formatted query comprises determining a first predicted intent vector to the embedding formatted query, wherein the first predicted intent vector comprises first confidence scores for all possible intents for which the first machine-learning logic model has been trained in response to the embedding formatted query; and determining, by the second machine-learning logic model, the second predicted intent to the embedding formatted query comprises determining a second predicted intent vector to the embedding formatted query, wherein the second predicted intent vector comprises second confidence scores for all possible intents for which the second machine-learning logic model has been trained in response to the embedding formatted query. ### Claim 10 (depends on 9) The method of claim 9, wherein combining the first predicted intent and the second predicted intent into the feature vector comprises: combining the first predicted intent vector and the second predicted intent vector into the feature vector that comprises a listing of all possible intents for which the first machine-learning logic model and the second machine-learning logic model have been trained with associated confidence scores ranking the possible intents as the query response. ### Claim 11 A system comprising: one or more processors and non-transitory computer-readable media having stored therein instructions, which when executed by the one or more processors, causes the system to perform functions comprising: receiving, from a computing device, a natural language-based query; converting the natural language-based query into an embedding formatted query having a vector format; inputting the embedding formatted query into a machine-learning logic model; determining, by the machine-learning logic model, a query response to the embedding formatted query that is predicted by the machine-learning logic model to be in context of beauty or cosmetic industry, wherein the machine-learning logic model is trained using a multi-dimensional learned-embedding that includes semantically similar terms in proximity in an embedding space, wherein the embedding space is limited to salient terms associated with the beauty or cosmetic industry, wherein determining, by the machine-learning logic model, the query response to the embedding formatted query comprises: determining, by a first machine-learning logic model, a first predicted intent to the embedding formatted query, wherein the first predicted intent is predicted by the first machine-learning logic model to be in context of beauty or cosmetic industry, wherein the first machine-learning logic model is trained using data limited to salient terms associated with the beauty or cosmetic industry; determining, by a second machine-learning logic model, a second predicted intent to the embedding formatted query, wherein the second predicted intent is different than the first predicted intent, wherein the second predicted intent is predicted using the second machine-learning logic model that is trained using data not limited to salient terms associated with the beauty or cosmetic industry; combining the first predicted intent and the second predicted intent into a feature vector; and processing, by the machine-learning logic model, the feature vector to determine the query response; and communicating, by the machine-learning logic model, the query response to the computing device. ### Claim 12 (depends on 11) The system of claim 11, wherein the first machine-learning logic model includes a first level model and a second level model, and the functions further comprise: training the first level model in the first machine-learning logic model using first level categories; and training the second level model in the first machine-learning logic model using second level categories more narrow than the first level categories. ### Claim 13 (depends on 11) The system of claim 11, wherein: determining, by the first machine-learning logic model, a first predicted intent vector to the embedding formatted query, wherein the first predicted intent vector comprises first confidence scores for all possible intents for which the first machine-learning logic model has been trained in response to the embedding formatted query; and determining, by the second machine-learning logic model, a second predicted intent vector to the embedding formatted query, wherein the second predicted intent vector comprises second confidence scores for all possible intents for which the second machine-learning logic model has been trained in response to the embedding formatted query. ### Claim 14 (depends on 13) The system of claim 13, wherein the functions further comprise: combining the first predicted intent vector and the second predicted intent vector into a feature vector that comprises a listing of all possible intents for which the first machine-learning logic model and the second machine-learning logic model have been trained with associated confidence scores ranking the possible intents as the query response; and processing, by the machine-learning logic model, the feature vector to determine the query response. ### Claim 15 A non-transitory computer-readable media having stored therein executable instructions, which when executed by a system one or more processors causes the system to perform functions comprising: receiving, from a computing device, a natural language-based query; converting the natural language-based query into an embedding formatted query having a vector format; inputting the embedding formatted query into a machine-learning logic model; determining, by the machine-learning logic model, a query response to the embedding formatted query that is predicted by the machine-learning logic model to be in context of beauty or cosmetic industry, wherein the machine-learning logic model is trained using a multi-dimensional learned-embedding that includes semantically similar terms in proximity in an embedding space, wherein the embedding space is limited to salient terms associated with the beauty or cosmetic industry, wherein determining, by the machine-learning logic model, the query response to the embedding formatted query comprises: determining, by a first machine-learning logic model, a first predicted intent to the embedding formatted query, wherein the first predicted intent is predicted by the first machine-learning logic model to be in context of beauty or cosmetic industry, wherein the first machine-learning logic model is trained using data limited to salient terms associated with the beauty or cosmetic industry; determining, by a second machine-learning logic model, a second predicted intent to the embedding formatted query, wherein the second predicted intent is different than the first predicted intent, wherein the second predicted intent is predicted using the second machine-learning logic model that is trained using data not limited to salient terms associated with the beauty or cosmetic industry; combining the first predicted intent and the second predicted intent into a feature vector; and processing, by the machine-learning logic model, the feature vector to determine the query response; and communicating, by the machine-learning logic model, the query response to the computing device. ### Claim 16 (depends on 15) The non-transitory computer-readable media of claim 15, wherein: determining, by the first machine-learning logic model, a first predicted intent vector to the embedding formatted query, wherein the first predicted intent vector comprises first confidence scores for all possible intents for which the first machine-learning logic model has been trained in response to the embedding formatted query; and determining, by the second machine-learning logic model, a second predicted intent vector to the embedding formatted query, wherein the second predicted intent vector comprises second confidence scores for all possible intents for which the second machine-learning logic model has been trained in response to the embedding formatted query.