# US 12,602,733 B2 · Automated machine learning-based topic tagging system for educational content categorization > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,602,733 B2 - Document ID: US12602733B2 - Status: Granted - Kind: B2 - Application number: 18/792,307 - Filed: 2024-08-01 - Published: 2026-02-05 - Granted: 2026-04-14 - Earliest claimed: 2024-08-01 - Assignees: Outschool, Inc. - Applicants: Outschool, Inc. - Inventors: Parima Shah; Brindusa Manolache; Christoph Lee; Axl Joshua Biol; Michael Brewer-Davis - Web page: https://patentgrep.com/patent/US12602733B2 ## CPC - G06Q 50/205: {Education administration or guidance} - G06F 16/35: Clustering; Classification - G06F 40/284: Lexical analysis, e.g. tokenisation or collocates - G06F 40/30: Semantic analysis - G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract The present invention provides a machine learning-based system designed to automatically generate and associate classes or videos offered on an educational platform with topic tags. The machine learning-based system leverages both extractive and predictive topic tagging techniques: extractive tagging identifies relevant topics directly from content, while predictive tagging uses machine learning models trained on pre-tagged content to predict relevant topics for new content. This dual approach ensures both accuracy and comprehensiveness in tagging. ## Claims ### Claim 1 An online education platform for tagging and storing activities, comprising: a tagged activities database comprising a plurality of activities with each activity tagged with one or more topics; a topics taxonomy database that stores a taxonomy of topics; one or more servers configured to remove low value words from a text, standardize the text into a desired format, combine words and/or phrases into a single string, calculate a similarity score between an embedding for the single string and an embedding for each topic in the taxonomy, select a plurality of topics in the topics taxonomy database with the highest similarity score, retain classified topics determined to have a high level of relevance, and tag an activity with each of the retained classified topics classified as having a strong match; a tokenizer configured to tokenize the text into the words and/or phrases; and a web interface: configured to interface with web devices of a plurality of users and receive the activity and the text from one of the plurality of users, configured to interface with an embedding model to determine a first embedding for the single string in the embedding model and determine second embeddings for each topic in the taxonomy of topics, wherein the embedding model comprises a neural network architecture or an advanced transformer-based model, wherein the first embedding is a first numerical value representing a first location of the single string in the embedding model and the second embeddings are second numerical values representing locations of the each topic in the taxonomy, configured to cause the one or more servers to calculate similarity scores between the first embedding and the second embeddings, configured to use the similarity scores to identify the plurality of topics, and configured to further refine the plurality of topics into the retained classified topics using a machine learning model that classifies each of the plurality of topics as having either a low, a middle or the high level of relevance, wherein the machine learning model comprises a processing unit, a memory and a neural network that comprises interconnected nodes organized in layers. ### Claim 2 (depends on 1) The online education platform of claim 1, wherein the activity is a class. ### Claim 3 (depends on 1) The online education platform of claim 1, wherein the activity is a video. ### Claim 4 (depends on 1) The online education platform of claim 1, wherein the similarity score is a cosine similarity score. ### Claim 5 (depends on 1) The online education platform of claim 1, wherein the machine learning model is an artificial intelligence model. ### Claim 6 (depends on 1) The online education platform of claim 1, wherein the machine learning model is a Generative Pre-trained Transformer (GPT) model. ### Claim 7 (depends on 1) The online education platform of claim 1, wherein the low, the middle and the high level of relevance comprise “Definitely Not Relevant,” “Not Sure” and “Definitely Relevant” respectively. ### Claim 8 (depends on 7) The online education platform of claim 7, wherein the web interface classifies each of the retained classified topics determined to have a high level of relevance as being either a bad, a weak or the strong match based at least in part on the text and relation to the activity. ### Claim 9 An online education platform for tagging and storing activities, comprising: a tagged activities database comprising a plurality of activities with each activity tagged with one or more topics; a topics taxonomy database that stores a taxonomy of topics; one or more servers configured to: remove low value words from a text, standardize the text into a desired format, combine words and/or phrases into a single string, access an embedding model to calculate a similarity score between an embedding for the single string and an embedding for each topic in the taxonomy, select a plurality of topics in the taxonomy database with the highest similarity score, using a machine learning model to retain classified topics determined to have a high level of relevance from the plurality of topics in the taxonomy database with the highest similarity score, and tag an activity with each of the retained classified topics classified as having a strong match; a tokenizer configured to tokenize the text into the words and/or phrases; a web interface configured to interface with web devices of a plurality of users and receive the activity and the text from one of the plurality of users; wherein the machine learning model comprises a processing unit, a memory and a neural network that comprises interconnected nodes organized in layers, configured for classifying each of the selected plurality of topics as having either a low, a middle or the high level of relevance, classifying each of the retained classified topics determined to have a high level of relevance as being either a bad, a weak or the strong match based at least in part on a text and relation to the activity; and the embedding model is configured for embedding the single string and embedding each topic in a taxonomy, wherein the embedding model comprises a neural network architecture or an advanced transformer-based model. ### Claim 10 (depends on 9) The online education platform of claim 9, wherein the activity is a class. ### Claim 11 (depends on 9) The online education platform of claim 9, wherein the activity is a video. ### Claim 12 (depends on 9) The online education platform of claim 9, wherein the similarity score is a cosine similarity score. ### Claim 13 (depends on 9) The online education platform of claim 9, wherein the machine learning model is an artificial intelligence model. ### Claim 14 (depends on 9) The online education platform of claim 9, wherein the machine learning model is a Generative Pre-trained Transformer (GPT) model. ### Claim 15 (depends on 9) The online education platform of claim 9, wherein the low, the middle and the high level of relevance comprise “Definitely Not Relevant,” “Not Sure” and “Definitely Relevant” respectively. ### Claim 16 (depends on 15) The online education platform of claim 15, wherein the bad, the weak and the strong match comprise “Bad Match,” “Weak Match” and “Strong Match” respectively. ### Claim 17 A method for tagging an educational activity with one or more topics, comprising: receiving a text that corresponds to the educational activity; removing low value words from the text; standardizing the text into a desired format; tokenizing the text into words and/or phrases; combining the tokenized words and phrases into a single string; generating an embedding for the single string using an embedding model, wherein the embedding model comprises a neural network architecture or an advanced transformer-based model; generating an embedding for each topic in a taxonomy of topics; calculating a cosine similarity score between the embedding for the single string and the embedding for each topic in the taxonomy; selecting a plurality of topics in the taxonomy that have the highest cosine similarity scores with the single string; classifying each topic in the selected plurality of topics using an artificial intelligence (AI) model as “Definitely Not Relevant”, “Not Sure”, and “Definitely Relevant”, wherein the AI model comprises a processing unit, a memory and a neural network that comprises interconnected nodes organized in layers; retaining the classified topics that are determined to be “Definitely Relevant”; classifying each of the retained topics using an artificial intelligence model as “Bad Match”, “Weak Match” and “Strong Match” based on at least part of the text and relation to the educational activity; and applying the retained topics classified as “Strong Match” as the one or more tags for the educational activity. ### Claim 18 (depends on 17) The method of claim 17, wherein the text comprises a title of the activity, a description of the activity and one or more learning objectives of the activity. ### Claim 19 (depends on 17) The method of claim 17, wherein the educational activity is a class. ### Claim 20 (depends on 17) The method of claim 17, wherein the education activity is a video.