mexca.text.sentiment

Extract sentiment from text.

Module Contents

Classes

SentimentExtractor

Extract sentiment from text.

Functions

cli()

Command line interface for sentiment extraction.

class mexca.text.sentiment.SentimentExtractor(model_name: str | None = None, device: torch.device | None = None)

Extract sentiment from text.

Parameters:

• model_name (str, optional) – The name of the text sequence classification model on Hugging Face hub used for sentiment prediction. By default ‘cardiffnlp/twitter-xlm-roberta-base-sentiment’.

• device (torch.device, optional, default=None) – The device on which sentiment extraction is performed. If None, defaults to ‘cpu’.

tokenizer

The pretrained tokenizer for sequence classification. Loaded automatically from model_name.

Type:

transformers.PreTrainedTokenizer

property classifier: transformers.XLMRobertaForSequenceClassification

The pretrained sequence classification model for sentiment prediction. Loaded automatically from model_name.

apply(transcription: mexca.data.AudioTranscription, show_progress: bool = True) → mexca.data.SentimentAnnotation

Extract the sentiment from text.

Iterates over the sentences in the audio transcription and predicts the sentiment (negative, neutral, positive).

Parameters:

• transcription (AudioTranscription) – The transcription of the speech segments in the audio fie split into sentences. Returned by AudioTranscriber.

• show_progress (bool, optional, default=True) – Whether a progress bar is displayed or not.

Returns:

An data class object with the positive, negative, and neutral sentiment scores for each sentence.

Return type:

SentimentAnnotation

mexca.text.sentiment.cli()

Command line interface for sentiment extraction. See extract-sentiment -h for details.