Sentiment#

Basic#

The easiest way to add sentiment to a text field is using the ds.extract_sentiment function.

Prior to adding sentiment, we will need to make sure to install HuggingFace’s Transformers.

pip install -q transformers

from relevanceai import Client
client = Client()
ds = client.Dataset("sample")

# Easily switch to a different HuggingFace model
ds.extract_sentiment(
   text_fields=["sample_1_label"],
)

For every document, you will get functions and formulas similar to the ones below:

{
   "_sentiment_": {
      "sample_1_label": {
         {
             "model_name": {
             "sentiment": sentiment, # positive / neutral / negative
             "score": np.round(float(scores[ranking[0]]), 4), # confidence of the sentiment
             "overall_sentiment_score": score if sentiment == "positive" else -score,
             # an overall sentiment score where -1 is negative and +1 is positive
         }
      }
   }
}

API Reference#

Add Sentiment to your dataset

class relevanceai.operations.text.sentiment.sentiments.SentimentOps#

get_shap_values(text, sentiment_ind=2, max_number_of_shap_documents=None, min_abs_score=0.1)#: Get SHAP values

class relevanceai.operations.text.sentiment.sentiment_workflow.SentimentWorkflow#

Sentiment workflow

fit_dataset(dataset, input_field, output_field='_sentiment_', log_to_file=True, chunksize=20, workflow_alias='sentiment', notes=None, refresh=False, highlight=False, positive_sentiment_name='positive', max_number_of_shap_documents=None, min_abs_score=0.1)#: Fit on dataset

Vectorize