omariabdelilah/marrakech-tourism-llm

Data and code for a single-inference zero-shot LLM framework for multi-dimensional tourism review analysis # marrakech-tourism-llm Data and code for the paper: > **A Single-Inference Zero-Shot LLM Framework for Multi-Dimensional Analysis of Machine-Translated Multilingual Tourism Reviews** The paper presents a framework based on Large Language Models that extracts six analytical dimensions — sentiment polarity, sentiment intensity, emotional tone, aspect-level opinions, reported problems, and economic signals — from a single zero-shot inference, and applies it to 14,838 TripAdvisor reviews of two heritage attractions in Marrakech, Morocco (the Medina and Bahia Palace). This repository releases the extracted-signal dataset and the analysis code, so that the reported distributions, cross-site contrasts, temporal trends and cross-dimensional statistics can be recomputed from scratch. --- ## Repository layout ``` marrakech-tourism-llm/ ├── data/ │ ├── llm_extracted_signals.csv 14,838 rows — the released dataset │ └── DATA_DICTIONARY.md column-by-column description ├── code/ │ ├── 00_prepare_data.py builds the released dataset (see note below) │ ├── 01_analysis.py reproduces the tables and statistics │ ├── 02_figures.py reproduces the figures │ ├── 03_baselines.py VADER and BERT baselines (see note below) │ └── extraction_prompt.txt the zero-shot prompt, verbatim as executed ├── outputs/ written by the scripts ├── requirements.txt ├── LICENSE MIT — applies to the code └── LICENSE-DATA CC BY 4.0 — applies to data/ ``` ## Quick start ```bash git clone github.com /marrakech-tourism-llm.git cd marrakech-tourism-llm pip install -r requirements.txt python code/01_analysis.py # tables and statistics -> outputs/ python code/02_figures.py # figures -> outputs/ ``` Both scripts read only `data/llm_extracted_signals.csv` and take under a minute on a laptop. No API key, no GPU and no network access are required. ## …