File size: 6,065 Bytes
21357a8 6a601a0 21357a8 bba9d71 21357a8 6a601a0 21357a8 80e20d5 21357a8 6a601a0 21357a8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 |
import streamlit as st
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import requests
import pandas as pd
import altair as alt
from collections import OrderedDict
from nltk.tokenize import sent_tokenize
import trafilatura
# Load the punkt tokenizer from nltk
import nltk
nltk.download('punkt')
# Load model and tokenizer
model_name = 'dejanseo/sentiment'
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
# Sentiment labels as textual descriptions
sentiment_labels = {
0: "very positive",
1: "positive",
2: "somewhat positive",
3: "neutral",
4: "somewhat negative",
5: "negative",
6: "very negative"
}
# Background colors for sentiments
background_colors = {
"very positive": "rgba(0, 255, 0, 0.5)",
"positive": "rgba(0, 255, 0, 0.3)",
"somewhat positive": "rgba(0, 255, 0, 0.1)",
"neutral": "rgba(128, 128, 128, 0.1)",
"somewhat negative": "rgba(255, 0, 0, 0.1)",
"negative": "rgba(255, 0, 0, 0.3)",
"very negative": "rgba(255, 0, 0, 0.5)"
}
# Function to get text content from a URL
def get_text_from_url(url):
downloaded = trafilatura.fetch_url(url)
if downloaded:
return trafilatura.extract(downloaded)
return ""
# Function to classify text
def classify_text(text, max_length):
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=max_length)
with torch.no_grad():
outputs = model(**inputs)
scores = torch.nn.functional.softmax(outputs.logits, dim=-1)[0].tolist()
return scores
# Function to handle long texts
def classify_long_text(text):
max_length = tokenizer.model_max_length
# Split the text into chunks
chunks = [text[i:i + max_length] for i in range(0, len(text), max_length)]
aggregate_scores = [0] * len(sentiment_labels)
chunk_scores_list = []
for chunk in chunks:
chunk_scores = classify_text(chunk, max_length)
chunk_scores_list.append(chunk_scores)
aggregate_scores = [x + y for x, y in zip(aggregate_scores, chunk_scores)]
# Average the scores
aggregate_scores = [x / len(chunks) for x in aggregate_scores]
return aggregate_scores, chunk_scores_list, chunks
# Function to classify each sentence in the text
def classify_sentences(text):
sentences = sent_tokenize(text)
sentence_scores = []
for sentence in sentences:
scores = classify_text(sentence, tokenizer.model_max_length)
sentiment_idx = scores.index(max(scores))
sentiment = sentiment_labels[sentiment_idx]
sentence_scores.append((sentence, sentiment))
return sentence_scores
# Streamlit UI
st.title("Sentiment Classification Model by DEJAN")
url = st.text_input("Enter URL:")
if url:
text = get_text_from_url(url)
if text:
scores, chunk_scores_list, chunks = classify_long_text(text)
scores_dict = {sentiment_labels[i]: scores[i] for i in range(len(sentiment_labels))}
# Ensure the exact order of labels in the graph
sentiment_order = [
"very positive", "positive", "somewhat positive",
"neutral",
"somewhat negative", "negative", "very negative"
]
ordered_scores_dict = OrderedDict((label, scores_dict[label]) for label in sentiment_order)
# Prepare the DataFrame and reindex
df = pd.DataFrame.from_dict(ordered_scores_dict, orient='index', columns=['Likelihood']).reindex(sentiment_order)
# Use Altair to plot the bar chart
chart = alt.Chart(df.reset_index()).mark_bar().encode(
x=alt.X('index', sort=sentiment_order, title='Sentiment'),
y='Likelihood'
).properties(
width=600,
height=400
)
st.altair_chart(chart, use_container_width=True)
# Display each chunk and its own chart
for i, (chunk_scores, chunk) in enumerate(zip(chunk_scores_list, chunks)):
chunk_scores_dict = {sentiment_labels[j]: chunk_scores[j] for j in range(len(sentiment_labels))}
ordered_chunk_scores_dict = OrderedDict((label, chunk_scores_dict[label]) for label in sentiment_order)
df_chunk = pd.DataFrame.from_dict(ordered_chunk_scores_dict, orient='index', columns=['Likelihood']).reindex(sentiment_order)
chunk_chart = alt.Chart(df_chunk.reset_index()).mark_bar().encode(
x=alt.X('index', sort=sentiment_order, title='Sentiment'),
y='Likelihood'
).properties(
width=600,
height=400
)
st.write(f"Chunk {i + 1}:")
st.write(chunk)
st.altair_chart(chunk_chart, use_container_width=True)
# Sentence-level classification with background colors
st.write("Extracted Text with Sentiment Highlights:")
sentence_scores = classify_sentences(text)
for sentence, sentiment in sentence_scores:
bg_color = background_colors[sentiment]
st.markdown(f'<span style="background-color: {bg_color}">{sentence}</span>', unsafe_allow_html=True)
else:
st.write("Could not extract text from the provided URL.")
# Additional information at the end
st.markdown("""
Multi-label sentiment classification model developed by [Dejan Marketing](https://dejanmarketing.com/).
The model is designed to be deployed in an automated pipeline capable of classifying text sentiment for thousands (or even millions) of text chunks or as a part of a scraping pipeline. This is a demo model which may occassionally misclasify some texts. In a typical commercial project, a larger model is deployed for the task, and in special cases, a domain-specific model is developed for the client.
### Engage Our Team
Interested in using this in an automated pipeline for bulk sentiment processing?
Please [book an appointment](https://dejanmarketing.com/conference/) to discuss your needs.
""")
|