RashidNLP's picture
Update README.md
718cca4
|
raw
history blame
1.65 kB
metadata
datasets:
  - financial_phrasebank
  - chiapudding/kaggle-financial-sentiment
  - zeroshot/twitter-financial-news-sentiment
  - FinanceInc/auditor_sentiment
language:
  - en
library_name: transformers
tags:
  - Sentiment Classification
  - Finance
  - Deberta-v2

Deberta for Financial Sentiment Analysis

I use a Deberta model trained on over 1 million reviews from Amazon's multi-reviews dataset and finetune it on 4 finance datasets that are categorized with Sentiment labels. The datasets I use are

  1. financial_phrasebank
  2. chiapudding/kaggle-financial-sentiment
  3. zeroshot/twitter-financial-news-sentiment
  4. FinanceInc/auditor_sentiment

How to use the model

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

def get_sentiment(sentences):
    bert_dict = {}
    vectors = tokenizer(sentences, padding = True, max_length = 65, return_tensors='pt').to(device)
    outputs = bert_model(**vectors).logits
    probs = torch.nn.functional.softmax(outputs, dim = 1)
    for prob in probs:
        bert_dict['neg'] = round(prob[0].item(), 3)
        bert_dict['neu'] = round(prob[1].item(), 3)
        bert_dict['pos'] = round(prob[2].item(), 3)
        print (bert_dict)

MODEL_NAME = 'RashidNLP/Finance_Multi_Sentiment'
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

bert_model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME, num_labels = 3).to(device)
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

get_sentiment(["The stock market will struggle until debt ceiling is increased", "ChatGPT is boosting Microsoft's search engine market share"])