import spacy
from spacy import displacy
import gradio as gr
nlp = spacy.load("en_core_web_sm")
def text_analysis(text):
doc = nlp(text)
html = displacy.render(doc, style="dep", page=True)
html = (
"
"
+ html
+ "
"
)
pos_count = {
"char_count": len(text),
"token_count": 0,
}
pos_tokens = []
for token in doc:
pos_tokens.extend([(token.text, token.pos_), (" ", None)])
return pos_tokens, pos_count, html
iface = gr.Interface(
text_analysis,
gr.inputs.Textbox(placeholder="Enter sentence here..."),
["highlight", "key_values", "html"],
examples=[
["What a beautiful morning for a walk!"],
["It was the best of times, it was the worst of times."],
],
)
iface.test_launch()
if __name__ == "__main__":
iface.launch()