Contains LRGraph and EojeolCounter classes that build and query left-right substring frequency graphs from a corpus. This is a core data structure used across noun extraction, word extraction, and POS tagging. It is self-contained, requires only basic Python, and can be extracted for any substring frequency analysis task (e.g., collocation detection, morpheme segmentation). The implementation demonstrates non-obvious skill in efficient graph construction and traversal for unsupervised morphological analysis.

Provides functions to decompose and compose Hangul syllables into initial, medial, and final jamo (phonemes). This is a clean, self-contained utility that handles Unicode Hangul syllable blocks (U+AC00–U+D7A3). It can be reused in any Korean text processing project for tasks like normalization, sorting, or phonetic analysis. The implementation is efficient and well-documented, demonstrating deep understanding of the Hangul syllable structure.

Contains functions for normalizing Korean text: emoticon_normalize, repeat_normalize, only_hangle, and others. These are isolated, parameterized routines that solve common text cleaning problems (repeated characters, emoticon handling, filtering non-Hangul). They can be extracted and applied to any Korean text dataset with minimal adaptation. The code is concise and demonstrates practical data cleaning patterns.

Implements BaseVectorizer, a TF-IDF-like document vectorizer that supports fit-to-file for large corpora (streaming from disk) and encode/decode for bag-of-words representations. It is self-contained, parameterized (min_tf, max_tf, min_df, max_df), and can be reused for any text vectorization task. The fit-to-file pattern is a non-obvious optimization for memory-constrained environments, demonstrating good data engineering practice.