Skip to content

Finding correlations between human sentiment and machine sentiment through a tourist attraction reviews'

Notifications You must be signed in to change notification settings

annafonte/nlp-tripadvisor

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ironhack Logo

Extracting emotion through Machine Learning

Anna Fonte Farré Data Analytics, Barcelona October 2020

Content

Project Description

The development of Web 2.0 has led to an important amount of user generated content online. Users are now free to express their opinions about products, places and events. This project is aimed at introducing sentiment analysis into touristic attractions. To begin with, reviews from Sagrada Família were collected using a TripAdvisor scraper. Afterwards, two sentiment labels were created: the human sentiment which is the rate of the reviewer; and the machine sentiment which is extracted from the library NLTK. After that, classification models are built so as to predict polarity sentiments. Finally, a subgroup discovery analysis was performed so as to extract valuable information about negative reviews.

Questions & Hypotheses

The main objective of the project is creating a first attempt for Sentiment Analysis for tourist attractions, as mostly of the research has been only done in the hospitality industry.

Dataset

As noted above, the data has been collected by scrapping TripAdvisor with a driver. Arround 55K opinions have been collected from 2010 to 2020. The main information that contains the dataset is related to the date of the visit, the location of the reviewer, the review title and the review body.

Workflow

  • Firstly, some research was done in order to find interesting questions and get a solid background about the topic.
  • Then, data was collected using Selenium and ChromeDriver from TripAvisor website.
  • For labelling the data, the human sentiment and machine sentiment approached were considered, as explained before.
  • Afterwards, the data cleaning and wrangling was performed, adapting all the features from the dataset and its types.
  • To continue, some operations were in place in order to deal with the categorical features: the text was preprocessed in order to use the NLP methods.
  • Finally, analysis was conducted in order to find correlation between the two different label approaches and also to discover interesting patterns in the negative subgroup.

Organization

All the steps of the project were organized with Trello (find the link attached below). Regarding the repository, it contains three main folders: the first one with the data used in the project and the second contains the notebooks for data collection, data cleaning, data analysis & data preprocessing and modelling.

Links

Repository
Slides
Trello