We will use the model in『SiEBERT - English-Language Sentiment Classification』 to classify the sentiment of English sentences. I would like to classify the sentiment of each English sentence by positive and negative.
Hartmann, Jochen and Heitmann, Mark and Siebert, Christian and Schamp, Christina, “More than a feeling: Accuracy and Application of Sentiment Analysis”, International Journal of Research in Marketing(2022)
Here we use the siebert/sentiment-roberta-large-english model on huggingface, if you want to use transformers on Google Colab you need to install transformers beforehand.
import numpy as np
from transformers import pipeline
from IPython.display import HTML
sentiment_pipeline = pipeline(
"sentiment-analysis", model="siebert/sentiment-roberta-large-english"
)
The entire text is separated by “.” to separate each sentence.
Here, Petrobras Webcast – 3rd Quarter Results 2022 November 5, 2022 transcription data is used.
transcript = """Hello!Hello!Hello!Hello!Hello!"""
ts_list = [ts for ts in transcript.split(".") if len(ts) > 20]
scores = sentiment_pipeline(ts_list)
Visualize the results using positive and negative labels and their scores.
for t, s in zip(ts_list, scores):
score = np.round(float(s["score"]), 4) # sentiment score
font_weight = "bold" if score > 0.995 else "normal" # thickness of the text
# color display for each sentiment
if s["label"] == "NEGATIVE":
r = 255 - 10 * int(1000 - score * 1000)
display(
HTML(
f"[score={score}] <span style='color:rgb({r},100,100);font-weight:{font_weight};'>{t}</span>"
)
)
elif s["label"] == "POSITIVE":
g = 255 - 10 * int(1000 - score * 1000)
display(
HTML(
f"[score={score}] <span style='color:rgb(100,{g},100);font-weight:{font_weight};'>{t}</span>"
)
)
[score=0.9976] Hello!Hello!Hello!Hello!Hello!