Sentiment Detection Project

A joint project with my fellow student, a part of the course in Textual Data Analysis, is devoted to building a sentiment detection system that is a common task in Natural Language Processing. The project is composed of three milestones.

Milestone I

Including:

• training sentiment classifiers on two datasets coming from different domains
• reporting performance of classifiers in terms of accuracy/precision/recall/F-score on the respective test sets
• performing a small qualitative assessment of the mistakes made by the classifiers.

Milestone II

Including:

• testing classifiers on datasets opposite to ones used for training
• reproting gained results and drawing conclusions about the transferability of trained sentiment classifiers across different domains

Milestone III

Including:

• translating one of the datasets used for training the sentiment classifier in the Russian language
• training a new sentiment classifier on produced dataset
• evaluating gained results both qualitatively and quanititatively

Data used

1. Sentiment Labelled Sentences Data Set

The datasets contains sentences extracted from reviews of products, movies, and restaurants and labelled with positive or negative sentiment. The sentences are derived from three different websites

• imdb.com (movies reviews)
• amazon.com (product reviews, mostly mobile phones, headsets, and some other phone accessories)
• yelp.com (restaurants reviews)

For each website, there are 500 positive and 500 negative sentences that ensure the dataset to be balanced. Altogether the dataset consists of 3,000 instances. Thus it's a medium sized which allows to browse the whole dataset as needed.

2. IMDB reviews (Large Movie Review Dataset)

The dataset provides sufficient data for both positive and negative sentiments which are uniformly distributed (label-balanced). Length of individual reviews is longer (than a single sentence like in Sentiment Labelled Sentences Data Set). Already known in NLP & ML community. Suitable for binary sentiment classification.

Please head over to data folder for datasets used in the project.

Packages used

• python 3.7.3
• re 2.2.1
• numpy 1.16.3
• pandas 0.24.2
• matplotlib 3.0.3
• seaborn 0.9.0
• scikit-learn 0.20.3
• eli5 0.8.2