top of page

Social Media Listening: Malaysian
Socio-Economic
Sentiment Analysis

CHALLENGE

Public sentiment during national transitions is highly fragmented and noisy. Government agencies need structured, real-time feedback on policy transitions rather than manual, time-consuming forum audits.

SOLUTION

Automating the categorization of unstructured Facebook and Twitter commentary into actionable national topics.

VALUE
  • ​It provides government agencies with a real-time "temperature check" on how the public reacts to major national policies.

  • By segmenting and analyzing the pillar, policymakers can pinpoint exactly when and where grocery inflation complaints are spiking on social media.

  • Social media is a massive, untapped goldmine of public opinion. This pipeline converts thousands of chaotic, disorganized posts into a clean, queryable relational database ready to power interactive BI Dashboards (like Tableau or Power BI).

Limitations
  • Cross-Linguistic Sentiment Drift & Neutrality Bias
    Using TextBlob (built purely on English lexicons) to analyze Malay text causes unrecognized vocabulary to default to a neutral 0.0 polarity score. 

    RecommendationReplace lexicon-based scoring with a dedicated Bahasa Malaysia NLP library or a pretrained multilingual deep learning model. This transitions the pipeline to true machine learning, enabling contextual sentiment capture of code-mixed local dialects (Manglish/Malay).

     

  • Static, Rule-Based Topic Modeling
    Sorting posts into categories relies strictly on hardcoded Regular Expressions (Regex) keywords. The pipeline is blind to emerging, unprogrammed public concerns or shifts in slang terms.

    RecommendationImplement a hybrid architecture using BERTopic alongside rule-based constraints. While Regex maintains precision on target policy topics, BERTopic leverages sentence embeddings in the background to dynamically cluster and auto-discover unexpected, trending conversation pillars.

     

  • Morphological Preprocessing Mismatch
    Applying English tokenizers and stemmers fails on Malay grammar, which relies on complex affixes.

    RecommendationStandardize preprocessing using Malaya's native Malay stemmer and stopword list. This properly resolves morphological roots and filters noise according to local linguistic structures, ensuring correct token normalization before modeling.

Built a targeted, rule-based NLP pipeline to parse unstructured social media commentary, segmenting public opinion into key socio-economic pillars during major government transitions.

#NaturalLanguageProcessing
#TextAnalytics #RuleBased
#SentimentAnalysis #Python
#SocialMediaListening #TextbBlob
bottom of page