π€ Hybrid Multilingual RAG + Ensemble Sentiment + Economic Forecast
ENSSEA β Master's Thesis | Si Tayeb Houari | 2025β2026
No index built yet.
0 1
ποΈ Speak your question β get a spoken answer
π€ Hybrid Multilingual RAG Framework
| Component | Details |
|---|---|
| π« School | ENSSEA β Γcole Nationale SupΓ©rieure de Statistique et d'Γconomie AppliquΓ©e |
| π€ Author | Si Tayeb Houari |
| π Year | 2025β2026 |
| π Degree | Master's β Statistics & Foresight Economics |
π§ Models Used
- π¦ FinBERT (ProsusAI) β Financial sentiment (40%)
- π XLM-RoBERTa (CardiffNLP) β Multilingual sentiment (30%)
- π Economic Lexicon β Domain-specific keywords (30%)
- π MiniLM-L12 β Multilingual embeddings (FAISS)
- π ms-marco-MiniLM β Cross-encoder reranking
- π£οΈ Whisper-small β ASR
- π€ Llama-3.3-70B via Groq β Response generation
π Forecasting
- Baseline: ARIMA(1,1,1)
- Enhanced: SARIMAX + Ensemble Sentiment (n_test = 3)
- Tests: ADF, Granger Causality, Diebold-Mariano
- Data: World Bank API
π Economic Forecast β ARIMA vs SARIMAX + Ensemble Sentiment
n_test = 3 β Evaluates on the last 3 years (captures recent economic turbulence)
π― Target Variable
1990 2020
2010 2024