sohomghosh/LIPI_FinSim4_ESG_task2

How to use ths model? Download the pytorch_model.bin file and execute the following:

import pandas as pd
import torch
import transformers
from torch.utils.data import Dataset, DataLoader
from transformers import RobertaModel, RobertaTokenizer, BertModel, BertTokenizer

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

MAX_LEN = 128
BATCH_SIZE = 20
text_col_name = 'sentence'
category_col = 'label_text'

#Input should be one dataframe having one column with header as 'sentence' : test_df (do reset_index() if needed)
test_df = pd.DataFrame({"sentence":['We are striving to reduce the amount of waste we produce, and to reduce water as well as paper consumption.']})

def scoring_data_prep(dataset):
    out = []
    target = []
    mask = []
    
    for i in range(len(dataset)):
        rec = dataset[i]
        out.append(rec['ids'].reshape(-1,MAX_LEN))
        mask.append(rec['mask'].reshape(-1,MAX_LEN))

        out_stack = torch.cat(out, dim = 0)
        mask_stack = torch.cat(mask, dim =0 )
        out_stack = out_stack.to(device, dtype = torch.long)
        mask_stack = mask_stack.to(device, dtype = torch.long)

    return out_stack, mask_stack


class Triage(Dataset):
    """
    This is a subclass of torch packages Dataset class. It processes input to create ids, masks and targets required for model training. 
    """

    def __init__(self, dataframe, tokenizer, max_len, text_col_name):
        self.len = len(dataframe)
        self.data = dataframe
        self.tokenizer = tokenizer
        self.max_len = max_len
        self.text_col_name = text_col_name
        

    def __getitem__(self, index):
        title = str(self.data[self.text_col_name][index])
        title = " ".join(title.split())
        inputs = self.tokenizer.encode_plus(
            title,
            None,
            add_special_tokens=True,
            max_length=self.max_len,
            pad_to_max_length=True,
            return_token_type_ids=True,
            truncation=True,
        )
        ids = inputs["input_ids"]
        mask = inputs["attention_mask"]

        return {
            "ids": torch.tensor(ids, dtype=torch.long),
            "mask": torch.tensor(mask, dtype=torch.long),
            
        }

    def __len__(self):
        return self.len
        
class BERTClass(torch.nn.Module):
    def __init__(self, num_class):
        super(BERTClass, self).__init__()
        self.num_class = num_class
        self.l1 = RobertaModel.from_pretrained("roberta-base")
        self.pre_classifier = torch.nn.Linear(768, 768)
        self.dropout = torch.nn.Dropout(0.3)
        self.classifier = torch.nn.Linear(768, self.num_class)
        self.history = dict()

    def forward(self, input_ids, attention_mask):
        output_1 = self.l1(input_ids=input_ids, attention_mask=attention_mask)
        hidden_state = output_1[0]
        pooler = hidden_state[:, 0]
        pooler = self.pre_classifier(pooler)
        pooler = torch.nn.ReLU()(pooler)
        pooler = self.dropout(pooler)
        output = self.classifier(pooler)
        return output
        
def do_predict(model, tokenizer, test_df):
  test_set = Triage(test_df, tokenizer, MAX_LEN, text_col_name)
  test_params = {'batch_size' : BATCH_SIZE, 'shuffle': False, 'num_workers':0}
  test_loader = DataLoader(test_set, **test_params)
  out_stack, mask_stack = scoring_data_prep(dataset = test_set)
  n = 0
  combined_output = []
  model.eval()
  with torch.no_grad():
      while n < test_df.shape[0]:
          output = model(out_stack[n:n+BATCH_SIZE,:],mask_stack[n:n+BATCH_SIZE,:])
          n = n + BATCH_SIZE
          combined_output.append(output)
      combined_output = torch.cat(combined_output, dim = 0)
      preds = torch.argsort(combined_output, axis = 1, descending = True)
  preds = preds.to('cpu')
  actual_predictions = [i[0] for i in preds.tolist()]
  return actual_predictions
  
model_sustain = BERTClass(2)
model_sustain.to(device)
model_sustain.load_state_dict(torch.load('pytorch_model.bin', map_location=device)['model_state_dict'])

tokenizer_sus = BertTokenizer.from_pretrained('roberta-base')
actual_predictions_sus = do_predict(model_sustain, tokenizer_sus, test_df)

test_df['sustainability'] = ['sustainable' if i==0 else 'unsustainable' for i in actual_predictions_read]

Our work can be cited as follows:

@inproceedings{ghosh-2022-finsim-esg,
    title = "Ranking Environment, Social And Governance Related Concepts And Assessing Sustainability Aspect Of Financial Texts",
    author={Ghosh, Sohom and Naskar, Sudip Kumar},
    booktitle = "Proceedings of the Fourth Workshop on Financial Technology and Natural Language Processing (FinNLP@IJCAI-ECAI 2022)",
    month = "July" ,
    year = "2022",
    address = "Vienna, Austria",
    publisher = "-",
    url = "https://mx.nthu.edu.tw/~chungchichen/FinNLP2022_IJCAI/14.pdf",
    pages = "87--92",
}