Distributed Dictionary Representation

ddr performs semantic scoring of texts against dictionary categories using cosine similarity in embedding space, following the Distributed Dictionary Representation (DDR) method (Garten et al., 2018). This captures semantic proximity even when exact dictionary words are absent from the text.

Pass a gensim model name to automatically download embeddings (requires pip install liwca[ddr]):

results = liwca.ddr(texts, dx, "glove-wiki-gigaword-100")

Or bring your own embeddings as a dict-like mapping:

results = liwca.ddr(texts, dx, my_embeddings)

Values are cosine similarities in [-1, 1]. See ddr for full parameter details.