CLI Reference¶
The package installs the implicit-word-network command.
build¶
Build a network and write it to a file. The output format is inferred from
the file extension: .json (nodes + edges), .csv (edge list), .graphml /
.gexf (NetworkX, optionally with term nodes) or .npz (full network state
for ImplicitNetwork.load).
implicit-word-network build documents.txt -o network.json
implicit-word-network build documents.csv -o network.graphml \
--text-column body --id-column doc_id \
--extractor gliner --model gliner-community/gliner_medium-v2.5 \
--labels person organization location --threshold 0.4 --device cuda \
--window 2 --decay exponential --min-weight 0.5 --include-terms
implicit-word-network build documents.txt -o network.npz --extractor gazetteer --gazetteer gaz.json
Input files: plain text (one document per non-empty line) or CSV/TSV with a text column.
| Option | Description |
|---|---|
--extractor {spacy,gliner,gazetteer} |
Entity extractor (default spacy) |
--model |
spaCy pipeline or GLiNER model id |
--labels L [L ...] |
Entity types to keep / zero-shot labels |
--threshold |
GLiNER confidence threshold |
--gazetteer FILE |
JSON {label: [surface forms]} for the gazetteer extractor |
--device |
GLiNER device (cpu, cuda, mps) |
--window |
Context window in sentences (default 2) |
--decay |
exponential, constant, inverse, linear |
--include-stopwords |
Keep stop words as term nodes |
--min-weight, --top-k |
Edge filters for the export |
--include-terms |
Add term nodes (GraphML/GEXF) |
--batch-size |
Documents per extractor call |
-q, --quiet |
Hide progress bars |
summary¶
Build a network and print statistics, the most frequent entities and the strongest edges: