Novel Mining · 小说对话挖掘工作流

To extract structured dialogue from a novel text, this Skill parses raw narrative volumes and converts each utterance into JSON with speaker, text, and location metadata. This structured output serves as a foundation for building role-driven conversational datasets.

Dialogue addressee inference identifies the intended recipient of each utterance using rule-based multi-turn reasoning. This process analyzes surrounding conversation context to accurately map interaction patterns and build relationship matrices for multi-character systems.

Computing conditional response probabilities from a novel corpus involves a statistics engine that calculates how likely one character responds to another. It generates probability matrices and silence thresholds to support downstream chat triggering logic.

The dialogue mining workflow is designed for the Haruhi Suzumiya novel corpus across all narrative volumes. While the extraction logic applies rule-based inference, adapting it to other novels requires ensuring the raw text format matches the expected input structure.

You do not need external Python libraries to run the novel mining pipeline. The Skill relies on Python standard-library-friendly rule inference for addressee detection and processes raw text files to generate JSON annotation outputs and probability results.

Rule-based addressee detection in narrative text relies on standard-library inference without external NLP models, which may limit accuracy in highly ambiguous multi-character scenes. It processes explicit contextual cues to infer intended recipients and quantify attitude traits.