Trustpilot review analysis
1. Web Scraping
To begin, you need a site to run analysis on. For the purposes of this demonstration, I have used Upwork, a site for freelancers. I then run bespoke Python code against Trustpilot, using Pandas, BeautifulSoup & Numpy packages, scrape the reviews, rating, user, and date of review from Trustpilots site.
2. Polarity & Subjectivity
The next step is to analyse the text of the reviews themselves. For the purposes of this, I wanted to better understand the Polarity and Subjectivity of the reviews.
Polarity: On a -1 to 1 scale, how positive (1) or negative (-1) was the text in the review.
Subjectivity: On a 0 to 1 scale, how objective (0) or subjective (1) was the text in the review.
This allows us to quantify both the sentiment and the emotional aspect of the review itself.
3. Topic Modelling
The last bit of Python modelling was topic modelling. For this, I leveraged the scikit learn package. This then extracted 3 distinct "topics" from the reviews, and 10 words that contribute to each topic. For example, the most prevalent topic for Upwork was support, and some of the 10 words associated with support topic reviews were freelancer, customer, support. This to me suggests that a large portion of negativity to Upwork was more based around customer support offered, as opposed to issues with the platform itself. A business that understands this source of negativity can make proactive decisions to make the customer experience better in the future.