Tourism Comment Text Mining and Recommendation System Using NLP and Big Data

The project addresses the need for personalized tourism recommendations by mining user comments from Qunar's travel division, aiming to improve decision‑making, satisfaction, and provide actionable insights such as traffic, crowd levels, and suitability for families.

The workflow begins with full extraction of comment texts, followed by cleaning (deduplication, missing‑value handling, stop‑word removal), tokenization, and corpus training using word2vec.

Subsequent steps include keyword extraction with TF‑IDF, synonym expansion, sentiment classification, and scoring of both comments and attractions to produce ranked lists.

A case study on amusement‑park themes demonstrates how high‑frequency keywords are selected and refined into user‑friendly index terms, enhancing searchability.

The technical stack comprises PostgreSQL for storage, ElasticSearch for full‑text retrieval, Gensim for synonym extraction, SnowNLP for sentiment (noted as suboptimal), LDA for topic modeling, and Python‑based HTTP APIs, with Pandas for data analysis.

Key advantages of the architecture are fast query speed, high sentiment classification accuracy, comprehensive ranking, accurate recommendation reasons, low development cost, and rapid deployment.

The article concludes with reflections on the challenges of NLP, the need for deeper linguistic and deep‑learning expertise, and acknowledges the contributing team members.