note: we are currently improving documentation and fixing known env bugs - please be patient :)
HORUS is meta and multi-level framework designed to provide a set of features at word-level to boost natural language frameworks. It's architecure is based on image processing and text classification clustering algorithms and shows to be helpful especially to noisy data, such as microblogs.
We are currently investigating Named Entity Recognition (NER) as use case. This version supports the identification of classical named-entity types (LOC, PER, ORG).
java -XX:ParallelGCThreads=2 -Xmx2048m -jar /**X**/horus-ner/data/models/tweetnlp/ark-tweet-nlp-0.3.2.jar --model /**X**/horus-ner/data/models/tweetnlp/model.ritter_ptb_alldata_fixed.20130723
X = local path to the code
install CMake, Boost, Boost.Python, and X11
Ubuntu
OSX
Setup the python environment
run ./prerequisites_setup.sh
Setup the framework environment
(*) - setup openCV 3.1: OSx users can benefit from anaconda, which provides a running version of OpenCV 3.1.0 (see more at #issue 6)
python main.py --text="whitney houston has been honored in nyc"
python main.py --file="sentences.txt" --ds_format=0
python main.py --file="ritter_ner.tsv" --ds_format=1 --output_file="metadata" --output_format="json"
1. to process an input file (e.g., conll) and generate the horus feature file format
- examples/process_input_file.py
2. to convert a horus input file format to conll
- horus/core/data_conversion.py
3. to run some benchmarks
- experiments/benchmarking.py
export FLASK_APP=rest.py
user$ flask run
browser:
http://localhost:5000/annotate?text=paris hilton
pip install horus
import horus
config = HorusConfig()
sentence = "paris hilton was once the toast of the town"
extractor = FeatureExtraction(config, load_sift=1, load_tfidf=1, load_cnn=0, load_topic_modeling=1)
print(extractor.extract_features_text(sentence))