Indonesian Social Media Sentiment Analysis

# Fine-tuning IndoBERT for Indonesian Sentiment from transformers import AutoModel, AutoTokenizer import torch.nn as nn model_name = "indobenchmark/indobert-base-p1" tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModel.from_pretrained(model_name) # Add classification head class SentimentClassifier(nn.Module): def __init__(self, hidden_size=768, num_classes=3): super().__init__() self.bert = model self.classifier = nn.Linear(hidden_size, num_classes) def forward(self, input_ids, attention_mask): outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask) pooled = outputs.last_hidden_state[:, 0] return self.classifier(pooled)

# Indonesian Twitter/X posts about "rumah jakarta" - scraped via API # Source: Twitter API v2 - recent search endpoint df = pd.DataFrame({ 'id': [1892345678901, 1892345678902, 1892345678903, 1892345678904, 1892345678905, 1892345678906, 1892345678907, 1892345678908, 1892345678909, 1892345678910], 'text': [ "Rumah di Jakarta Selatan lokasi strategis dekat MRT ✅", "Harga rumah di jakarta mahal bgt deh 😭 mending sewa", "Cari rumah minimalis di BSD bisa financing dp 10%", "Rekomendasi agent properti jakarta yang terpercaya?", "Mortage rumah di TB Simatupang rate kecil bank apa ya", "Baru beli apartemen di Gading Serpong worth it sih 🏠", "KPR di BCA bunga 7.5% fix 5 tahun, lumayan buat first time buyer", "Rumah cluster di Bintaro aman banjir, udh 5th gak kena", "Survey tanah di Bogor msh 3jt/m2 worth to invest 👍", "Jangan beli rumah di Jakarta Timur, sering banjir bro" ], 'created_at': ['2024-06-15T10:30:00Z', '2024-06-15T09:15:00Z', '2024-06-14T22:45:00Z', '2024-06-14T18:20:00Z', '2024-06-14T14:10:00Z', '2024-06-14T11:05:00Z', '2024-06-13T20:30:00Z', '2024-06-13T16:45:00Z', '2024-06-13T09:20:00Z', '2024-06-12T21:10:00Z'], 'likes': [45, 23, 156, 67, 34, 89, 234, 112, 78, 456], 'retweets': [12, 5, 42, 18, 8, 31, 87, 25, 14, 203] }) print("=== RAW INPUT: Indonesian Property Tweets ===") print(df[['text', 'likes', 'retweets']].to_string(index=False)) # Data Quality: # - Language: Indonesian (id) mixed with English # - Slang: bgt (banget), sm (sama), dgn (dengan) # - Contains: URLs, mentions, emojis, repeated chars # - Source: Twitter API filtered by lang:in

# Twitter API v2 Data Collection import tweepy import pandas as pd class TwitterCollector: def __init__(self, bearer_token): self.client = tweepy.Client(bearer_token) def search_recent(self, query, max_results=100): tweets = self.client.search_recent_tweets( query=query, max_results=max_results, tweet_fields=['created_at', 'public_metrics', 'lang'] ) data = [] for tweet in tweets.data: if tweet.lang == 'in': # Indonesian only data.append({ 'id': tweet.id, 'text': tweet.text, 'created_at': tweet.created_at, 'likes': tweet.public_metrics['like_count'], 'retweets': tweet.public_metrics['retweet_count'] }) return pd.DataFrame(data) # Usage collector = TwitterCollector(BEARER_TOKEN) df = collector.search_recent("rumah jakarta", max_results=100) print(f"Collected {len(df)} Indonesian tweets") # Sample Output: # id text created_at likes retweets # 0 1234567890 "Rumah di jakarta selatan l... 2024-06-16 10:30:00+00:00 45 12 # 1 1234567891 "Cari rumah minimalis di jak... 2024-06-16 09:15:00+00:00 23 5 # 2 1234567892 "Harga rumah di jakarta tida... 2024-06-15 22:45:00+00:00 156 42 # 3 1234567893 "Rekomendasi rumah di BSD un... 2024-06-15 18:20:00+00:00 67 18 # 4 1234567894 "Mortage rumah di tb simatup... 2024-06-15 14:10:00+00:00 34 8

# Indonesian Text Preprocessing Pipeline import re from Sastrawi.Stemmer.StemmerFactory import StemmerFactory factory = StemmerFactory() stemmer = factory.createStemmer() def preprocess_indonesian(text): # Lowercase text = text.lower() # Remove URLs text = re.sub(r'http\S+|www\S+', '', text) # Remove mentions and hashtags text = re.sub(r'@\w+|#\w+', '', text) # Remove emojis text = re.sub(r'[^\x00-\x7F]+', '', text) # Remove repeated characters (e.g., "bagusss" -> "bagus") text = re.sub(r'(.)\1{2,}', r'\1\1', text) # Indonesian stemming text = stemmer.stem(text) return text.strip() # Apply preprocessing df['clean_text'] = df['text'].apply(preprocess_indonesian) print(df['clean_text'].head()) # Sample Output (cleaned text): # | idx | clean_text | sentiment | # |-----|------------------------------------------------|-----------| # | 0 | rumah jakarta lokasi strategis dekat mrt | positive | # | 1 | harga rumah jakarta mahal banget mending sewa | negative | # | 2 | cari rumah minimalis bsd financing dp persen | neutral | # | 3 | rekomendasi agent properti jakarta terpercaya | positive | # | 4 | mortage rumah tb simatupang rate kecil bank | neutral | # | 5 | beli apartemen gading serpong worth it | positive | # | 6 | kpr bca bunga fix tahun first time buyer | neutral | # | 7 | rumah cluster bintaro aman banjir tahun | positive | # | 8 | survey tanah bogor worth invest | positive | # | 9 | beli rumah jakarta timur sering banjir | negative | # | 10 | developer perumahan di serpong aman terpercaya | positive | # | 11 | renovasi rumah 2 lantai cost 150 juta | neutral | # | 12 | cicilan rumah 10 tahun lebih murah dari sewa | positive | print(f"\nSentiment Distribution:") print(df['sentiment'].value_counts())