Upload 5 files
Browse files
bert-sentiment-detection.ipynb
ADDED
The diff for this file is too large to render.
See raw diff
|
|
dataset/reviews.csv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
streamlit_app.py/Homepage.py
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
|
3 |
+
st.set_page_config(
|
4 |
+
page_title="Sentiment Detection"
|
5 |
+
)
|
6 |
+
|
7 |
+
st.title("Sentiment Detection")
|
8 |
+
|
9 |
+
st.sidebar.success("Select a page above.")
|
10 |
+
|
11 |
+
st.header("The Need for Sentiment Detection")
|
12 |
+
st.text("""
|
13 |
+
Spam detection algorithms are used to detect and filter junk and spam emails with a high level of accuracy.
|
14 |
+
It is said that around half of all emails are spam, depending on the user. These emails can include scams or viruses intended to cause harm.
|
15 |
+
""")
|
16 |
+
|
17 |
+
st.header("Data Source")
|
18 |
+
st.text("""
|
19 |
+
Data Source: Preprocessed TREC 2007 Public Corpus Dataset.
|
20 |
+
Link: https://www.kaggle.com/datasets/imdeepmind/preprocessed-trec-2007-public-corpus-dataset
|
21 |
+
|
22 |
+
""")
|
streamlit_app.py/pages/SentimentDetection.py
ADDED
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from os import path
|
2 |
+
import streamlit as st
|
3 |
+
|
4 |
+
# import pickle
|
5 |
+
|
6 |
+
# from tensorflow import keras
|
7 |
+
import tensorflow as tf
|
8 |
+
import torch
|
9 |
+
from torch import nn
|
10 |
+
from transformers import BertModel, BertTokenizer
|
11 |
+
|
12 |
+
|
13 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
14 |
+
MODEL_NAME = 'bert-base-cased'
|
15 |
+
|
16 |
+
# Build the Sentiment Classifier class
|
17 |
+
class SentimentClassifier(nn.Module):
|
18 |
+
|
19 |
+
# Constructor class
|
20 |
+
def __init__(self, n_classes):
|
21 |
+
super(SentimentClassifier, self).__init__()
|
22 |
+
self.bert = BertModel.from_pretrained(MODEL_NAME)
|
23 |
+
self.drop = nn.Dropout(p=0.3)
|
24 |
+
self.out = nn.Linear(self.bert.config.hidden_size, n_classes)
|
25 |
+
|
26 |
+
# Forward propagaion class
|
27 |
+
def forward(self, input_ids, attention_mask):
|
28 |
+
_, pooled_output = self.bert(
|
29 |
+
input_ids=input_ids,
|
30 |
+
attention_mask=attention_mask,
|
31 |
+
return_dict=False
|
32 |
+
)
|
33 |
+
# Add a dropout layer
|
34 |
+
output = self.drop(pooled_output)
|
35 |
+
return self.out(output)
|
36 |
+
|
37 |
+
# from keras_preprocessing.sequence import pad_sequences
|
38 |
+
|
39 |
+
|
40 |
+
# def predict(ham_spam):
|
41 |
+
# model = load_model(r'test_HSmodel_r.h5')
|
42 |
+
# with open('tokenizer.pickle','rb') as handle:
|
43 |
+
# tokenizer = pickle.load(handle)
|
44 |
+
# tokenizer.fit_on_texts(ham_spam)
|
45 |
+
# x_1 = tokenizer.texts_to_sequences([ham_spam])
|
46 |
+
# x_1 = pad_sequences(x_1, maxlen=525)
|
47 |
+
# predictions = model.predict(x_1)[0][0]
|
48 |
+
# return predictions
|
49 |
+
|
50 |
+
MODEL_PATH = path.join(path.dirname(__file__), "bert_model.h5")
|
51 |
+
|
52 |
+
|
53 |
+
@st.cache_resource
|
54 |
+
def load_model_and_tokenizer():
|
55 |
+
model = SentimentClassifier(3)
|
56 |
+
model.load_state_dict(torch.load(MODEL_PATH, map_location=torch.device('cpu')))
|
57 |
+
model.eval()
|
58 |
+
return model, BertTokenizer.from_pretrained('bert-base-cased')
|
59 |
+
|
60 |
+
|
61 |
+
def predict(content):
|
62 |
+
model, tokenizer = load_model_and_tokenizer()
|
63 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
64 |
+
|
65 |
+
encoded_review = tokenizer.encode_plus(
|
66 |
+
content,
|
67 |
+
max_length=160,
|
68 |
+
add_special_tokens=True,
|
69 |
+
return_token_type_ids=False,
|
70 |
+
pad_to_max_length=True,
|
71 |
+
return_attention_mask=True,
|
72 |
+
return_tensors="pt",
|
73 |
+
)
|
74 |
+
|
75 |
+
input_ids = encoded_review["input_ids"].to(device)
|
76 |
+
attention_mask = encoded_review["attention_mask"].to(device)
|
77 |
+
|
78 |
+
output = model(input_ids, attention_mask)
|
79 |
+
_, prediction = torch.max(output, dim=1)
|
80 |
+
|
81 |
+
class_names = ["negative", "neutral", "positive"]
|
82 |
+
|
83 |
+
return class_names[prediction]
|
84 |
+
|
85 |
+
|
86 |
+
def main():
|
87 |
+
# giving a title to our page
|
88 |
+
st.title("Sentiment detection")
|
89 |
+
contents = st.text_area("Please enter reviews/sentiment/setences/contents:")
|
90 |
+
|
91 |
+
prediction = ""
|
92 |
+
|
93 |
+
# Create a prediction button
|
94 |
+
if st.button("Analyze Spam Detection Result"):
|
95 |
+
prediction = predict(contents)
|
96 |
+
|
97 |
+
if prediction < 0.5:
|
98 |
+
st.success(prediction)
|
99 |
+
elif prediction > 0.5:
|
100 |
+
st.success(prediction)
|
101 |
+
|
102 |
+
|
103 |
+
if __name__ == "__main__":
|
104 |
+
main()
|
streamlit_app.py/pages/bert_model.h5
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2f0e320ff87ab99bcb76ed153f14d6973dfa7bd1570d022c6d8bee1e496323e7
|
3 |
+
size 433339657
|