Feat/sentiment analysis #34
Conversation
| This model is trained on 124M tweets from January 2018 to December 2021, | ||
| and is finetuned for sentiment analysis. | ||
| It outputs a label - Neutral, Positive or Negative - and a score ranging | ||
| from 0 to 1 - 0 being the most negative and 1, the most positive. |
There was a problem hiding this comment.
I think you've mis-interpreted score here. Score is more akin to the confidence in the assessment.
I actually have mixed results with nlptown model as well - the same positive text in the screenshots only yield 1 star with a score of 0.76.
| create or replace function {{target.schema}}.udf_sentiment_analysis(text STRING, output INTEGER) | ||
| returns STRING |
There was a problem hiding this comment.
your arguments need revision
text STRING, model INTEGER or STRING [more on this later], output VARIANT (if you still want to differentiate between label and score)
and returns STRING, you have to make sure L32 has to cast as a string.
| model = "cardiffnlp/twitter-roberta-base-sentiment-latest" | ||
| -- model = 'nlptown/bert-base-multilingual-uncased-sentiment' | ||
|
|
||
| def sentiment_analysis(text, model_id, output='score'): |
There was a problem hiding this comment.
your function name (sentiment_analysis) and your handler on L8 have to be equivalent. you also should have the equivalent # of arguments here and in your UDF call.
|
function name and handler should be quivalent |
This macro iterates through a piece of text to return the overall sentiment of that text.
First, the macro pre-processes the text removing unnecessary punctuation and stopwords to help increase the accuracy of the model. Subsequently, using the transformers library it applies a sentiment analysis pipeline based on a pre-trained model that will return either a score or a label for the text.
Recommendation is to use the following popular models:
cardiffnlp/twitter-roberta-base-sentiment-latest:
(https://huggingface.co/cardiffnlp/twitter-roberta-base-sentiment-latest)
This model is trained on 124M tweets from January 2018 to December 2021, and is finetuned for sentiment analysis.
It outputs a label - Neutral, Positive or Negative - and a score ranging from 0 to 1 - 0 being the most negative and 1,
the most positive.
nlptown/bert-base-multilingual-uncased-sentiment:
(https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment)
This model is fine-tuned for sentiment analysis on product reviews in six languages: English, Dutch, German, French,
Spanish and Italian. It outputs a label - 1 to 5 stars - and a score ranging from 0 to 1 - 0 being the
most negative and 1, the most positive.
Macro returns a STRING data type. If 'score' is used as an output, then it will have to be cast to FLOAT data type.