Personalized learning systems and methods
OwnerZiv YIRMIYAHU, Oren LAADAN +1inventors
InventorsZiv YIRMIYAHU, Oren LAADAN, Giulia ROMEI
Life of this patent
Estimating the term…
Abstract
Systems, methods, and computer-readable media for personalized learning through content consumption. A learning platform receives content selected by or generated for a learner and processes it through a pipeline that includes content adaptation based on learner proficiency, speaker diarization, optimization for audio delivery, segmentation into manageable units, chunking into phrases using NLP-based linguistic analysis with hybrid LLM fallback and two-tier quality control, translation with phrase alignment using trace markers, and audio generation with accurate timestamp management.
Claims24
A computer-implemented system for personalized learning through content consumption, the system comprising:
- one or more processors;
- and
- one or more non-transitory computer-readable storage media storing instructions that, when executed by the one or more processors, cause the system to:
- receive, from a client device in communication with a learning platform, information identifying an item of content;
- process the item of content to generate a plurality of segments, wherein processing the item of content comprises adapting the item of content based on at least one of a proficiency of a user, a level of difficulty of the item of content, and a learning goal of the user;
- split each segment into a plurality of phrases based on linguistic analysis of the content, wherein a target phrase length is determined based on the proficiency of the user;
- generate, for each phrase, audio content in a foreign language and corresponding translated content in a native language of the user;
- and
- present, via the client device, the plurality of phrases to the user in an audio-first format that alternates between the foreign language and the native language.
A computer-implemented system for integrating micro-lessons into personalized learning content, the system comprising:
- one or more processors;
- and
- one or more non-transitory computer-readable storage media storing instructions that, when executed by the one or more processors, cause the system to:
- receive content that has been divided into a plurality of phrases;
- receive a set of learning objectives organized as a directed acyclic graph of dependencies with associated repetition requirements;
- for each learning objective, identify candidate insertion points within the plurality of phrases based on a computed fit between the learning objective and a context window of one or more adjacent phrases;
- select, from among the candidate insertion points, an optimal insertion point based on a value function that accounts for objective priority, repetition status, and context fit;
- and
- generate a micro-lesson at the selected insertion point, the micro-lesson comprising at least one of an adaptation of existing content and an extension of existing content with supplemental teaching material.
A computer-implemented method for quality-controlled content processing in a learning system, the method comprising:
- submitting, by one or more processors, a content processing task to a first language model;
- performing a syntactic quality check on an output of the first language model, the syntactic quality check comprising at least one of formatting validation, delimiter validation, and length validation;
- performing a semantic quality check on the output, the semantic quality check comprising at least one of a parts-of-speech comparison using natural language processing and a meaning preservation query using a language model;
- upon detecting a failure in at least one of the syntactic quality check and the semantic quality check, applying a cascading series of fallback methods comprising:
- attempting structural remediation to correct formatting errors;
- retrying the task with an adapted prompt in an agentic mode;
- retrying the task with a second language model that is more capable than the first language model;
- and
- falling back to deterministic NLP-based processing;
- and
- producing validated output for delivery to a learner.
A computer-implemented method for partial translation in a learning system, the method comprising:
- computing, by one or more processors, a personalized familiarity metric for each of a plurality of words in content to be presented to a learner, wherein the personalized familiarity metric is based on a combination of:
- a static familiarity component based on at least a base word frequency in a foreign language and a cognate similarity between the foreign language and a native language of the learner;
- a dynamic familiarity component based on at least a cumulative exposure history of the learner to each word and a temporal spacing of the exposures;
- and
- an assumed familiarity component based on a proficiency level of the learner;
- determining, based on the personalized familiarity metric for each word, whether to include or omit a translation of the word;
- and
- generating audio content for the learner that includes translations selectively based on the determining step.
Dependencies are inferred from the claim text.