Just a simple sentiment tool.

It just grabs a set of pre-made sentiment models that you can quickly use to attach sentiment scores to text. None of these sentiment models will be perfect, as none of them actually understand language, but they may serve well in human-in-the-loop kinds of labelling situations. Currently the tool only supports English models.

The goal is to whip up some sentiment models real quick. A demo is shown below.

import pandas as pd
from sentimany.sentiment import (
    vader_sentiment,
    textblob_sentiment,
    onnx_sentiment,
    roberta_sentiment,
    nlptown_sentiment,
)

# Add some text to a pandas dataframe
texts = [
    "i like dogs",
    "i hate cats",
    "stroopwafels are amazing",
    "mcdondals is horrible",
]
df = pd.DataFrame({"text": texts})

# Apply each sentiment model and attach it as a new column
(df
  .assign(vader = lambda d: vader_sentiment(d['text']), 
          textblob = lambda d: textblob_sentiment(d['text']),
          imdb_onnx = lambda d: onnx_sentiment(d['text'], "onnx/imdb-reviews.onnx"),
          amazon_onnx = lambda d: onnx_sentiment(d['text'], "onnx/amazon-reviews.onnx"),
          roberta = lambda d: roberta_sentiment(d['text']), 
          nlptown = lambda d: nlptown_sentiment(d['text'])))

This would result in a table that looks something like;

This tool is a bit of a hack, but you can install it via pip using git.

python -m pip install "sentimany @ git+https://github.com/koaning/sentimany.git"

If you'd like to use the pretrained onnx models (these are made with sklearn-onnx) you can download them manually from the onnx folder of this repo.

I like to build in public but I should stress that this is a repo made for utility for myself. Honestly, it's made in a quick evening. Feel free to re-use, but don't expect maintenance or production-quality code in the long term.

More-over though, keep in mind that sentiment models are imperfect and brittle. In particular check this short blogpost and this huggingface stream for more details.