System

A system comprising:

• circuitry configured to:
• acquire, from a terminal device coupled to a packet-switched network via a communication interface, character data representing a natural-language request and at least one of audio data and image data associated with a source of the character data;
• perform tokenization on the character data using a stored vocabulary to generate a token sequence, perform dependency parsing on the token sequence to generate a syntactic structure, and compute a semantic embedding of the token sequence using a neural network encoder;
• apply an intent classifier to the semantic embedding and features of the syntactic structure to generate request classification data comprising an intent label and extracted topic key phrases;
• retrieve, from a data storage device coupled to the packet-switched network, a user profile record associated with the source of the character data, the user profile record comprising attribute data, interaction history records, and a stored proficiency index;
• update the proficiency index by extracting, from the interaction history records, numerical features comprising interaction counts, topic difficulty distributions, clarification request frequencies, and temporal intervals, combining the numerical features into a feature vector, and applying a trained classification model to the feature vector;
• determine an affective state index by extracting acoustic features from the audio data and textual sentiment features from the character data, and applying a multi-modal classifier that fuses the acoustic features and the textual sentiment features to output a categorical affective state label;
• determine guidance parameter data specifying a complexity level, a content volume, an output style, and at least one of a stepwise explanation directive, an additional example directive, and a verification question directive, based on the intent label, the updated proficiency index, and the affective state index;
• retrieve, from a knowledge data region of the data storage device, reference content segments corresponding to the extracted topic key phrases and filtered by the complexity level;
• construct a prompt sequence for a generative neural network model by concatenating, in a structured template format, a role-specification segment, a context segment encoding the updated proficiency index and the affective state index, the guidance parameter data as instruction content, the reference content segments, and a query segment comprising the character data;
• encode the prompt sequence into a token index sequence using a subword tokenizer, transmit the token index sequence and generation condition parameters comprising a maximum output token count and a sampling temperature to the generative neural network model via the communication interface, and receive a generated token index sequence;
• decode the generated token index sequence into a natural-language output, apply content filtering using a binary classifier, perform vocabulary complexity adjustment by comparing tokens against a mapping table indexed by the updated proficiency index, and apply structural format conversion;
• transmit the post-processed natural-language output to the terminal device for presentation via at least one of visual display and audio output;
• and
• store the character data, the prompt sequence, and the post-processed natural-language output as a new interaction history record in the data storage device, and update the user profile record including the proficiency index and the affective state index for use in constructing subsequent prompt sequences with progressively adapted guidance parameter data.

A method performed by circuitry, the method comprising:

• acquiring, from a terminal device coupled to a packet-switched network via a communication interface, input data comprising at least one of audio data and character data;
• applying a natural language processing pipeline to the character data to extract topic identification data and request classification data;
• computing a proficiency index associated with a source of the input data by applying a trained machine learning model to feature vectors derived from interaction history records stored in a data storage device coupled to the packet-switched network;
• determining an affective state index by performing recognition processing on at least one of the audio data, the character data, and image data received from the terminal device;
• generating guidance parameter data specifying at least a complexity level, a content volume, and an output style based on the topic identification data, the proficiency index, and the affective state index;
• retrieving, from the data storage device, reference data corresponding to the topic identification data;
• constructing a prompt sequence for a generative neural network model by combining, in a structured template format, the guidance parameter data as instruction content, the reference data, and the request classification data;
• transmitting the prompt sequence to the generative neural network model via the communication interface, and receiving, from the generative neural network model, a natural-language output;
• applying post-processing to the natural-language output, the post-processing comprising at least one of content filtering by a classification model, vocabulary complexity adjustment based on the proficiency index, and structural format conversion;
• and
• storing the input data, the prompt sequence, and the post-processed natural-language output as an interaction history record in the data storage device, and updating the proficiency index based on the stored interaction history record for use in constructing a subsequent prompt sequence.