Information Processing Apparatus for Generative AI Response Filtering

The patent segments the response generation process into two independent stages: (1) a determination unit that first classifies the user request into target categories and decides whether to generate a response, and (2) a generation unit that only generates responses when the determination unit approves. This segmentation allows the system to maintain high generation speed for approved requests while preventing inappropriate responses through preliminary classification and determination.

The determination unit serves multiple functions simultaneously: it classifies user requests into target categories, determines whether the requests are appropriate for response generation, and controls the generation unit's operation. By making the determination unit multi-functional, the patent improves response appropriateness without proportionally increasing system complexity, as one component handles multiple tasks rather than adding separate dedicated components for each function.

Instead of trying to precisely classify all possible request types and their appropriate responses (the traditional approach), the patent inverts the logic by explicitly defining non-response targets - categories of requests that should NOT be responded to. The determination unit works by identifying what requests are inappropriate rather than exhaustively determining what requests are appropriate, which simplifies the classification task and improves both inappropriate response suppression and request classification precision.