System

A system comprising:

• circuitry configured to:
• receive, via a communication interface coupled to a packet-switched network, text data from a terminal device;
• perform normalization processing and structuring processing on the text data comprising applying subword segmentation to generate a token sequence, applying at least one of part-of-speech tagging or dependency parsing to the token sequence, and extracting an intent indicator and a condition indicator to generate internal representation data encoded as a key-value data structure;
• acquire, based on a user identifier, behavior history records from a storage device, perform aggregation processing comprising grouping the behavior history records by a plurality of attributes and computing recency-weighted preference scores, and integrate the preference scores with the internal representation data to generate user attribute information;
• construct a first message set comprising a system message and a user message embedding the text data, the internal representation data, and the user attribute information, invoke a generative neural network model using the first message set to generate a prompt sentence, and validate the prompt sentence by verifying presence of at least one term corresponding to the intent indicator;
• construct a second message set comprising the validated prompt sentence, invoke the generative neural network model using the second message set to generate response content, parse the response content to extract a plurality of candidate data records, compute similarity scores between attribute values in the candidate data records and attribute data in a reference data store, calculate ranking scores by combining the similarity scores with the preference scores, and select a subset of entries based on the ranking scores;
• and
• construct structured output data based on the selected subset and transmit the structured output data to the terminal device via the communication interface and the packet-switched network.

A method comprising:

• receiving, via a communication interface coupled to a packet-switched network, text data from a terminal device operated by a user;
• performing normalization processing on the text data comprising at least one of case conversion, control character removal, or whitespace standardization to generate normalized text data, and performing structuring processing on the normalized text data comprising applying subword segmentation to generate a token sequence and extracting, from the token sequence, an intent indicator and a condition indicator to generate internal representation data encoded as a key-value data structure;
• constructing a first message set comprising a system message defining a role constraint and a user message embedding the text data and the internal representation data, and invoking a generative neural network model using the first message set as input to generate a prompt sentence;
• constructing a second message set comprising the prompt sentence, invoking the generative neural network model using the second message set as input to generate response content, and parsing the response content by applying sentence segmentation and pattern recognition to extract a plurality of candidate data records each comprising an attribute field and a context field;
• computing similarity scores between attribute values in the candidate data records and attribute data stored in a reference data store, calculating ranking scores for entries in the reference data store by combining the similarity scores with preference weight values derived from history records associated with a user identifier stored in a storage device, and selecting a subset of entries having ranking scores exceeding a threshold value;
• and
• constructing structured output data based on the selected subset of entries and transmitting the structured output data to the terminal device via the communication interface and the packet-switched network.