Fake Review Detector
A small natural-language-processing program. It cleans raw review text, scores each review against a list of weighted keywords, and uses score thresholds to flag reviews as likely truthful or likely deceptive, writing the results to a report.
- Stack
- C++NLPFile I/OUnit Testing
- Highlights
- Preprocessing that strips punctuation and normalises case
- Weighted keyword scoring per review
- Threshold classification into truthful, deceptive or uncategorised
The problem
Fake reviews tend to lean on different words from genuine ones. Given a list of keywords weighted by how strongly they signal honesty or deception, read a set of reviews and sort them.
Approach
Clean the text first. Each review is split into words, then stripped of non-alphanumeric characters and lower-cased, so "Great!" and "great" count as the same keyword.
Score by weights. Every word is looked up in the keyword list, and a review's score is the sum of its words' weights. Words not in the list count as zero.
Classify and report. Scores above one threshold are marked truthful, and scores below another are marked deceptive. Everything in between stays uncategorised. The program writes each review's score and category to a report, then summary counts and the highest- and lowest-scoring reviews. Unit tests cover the scoring and preprocessing functions.