TrustPilotCalc

A package to calculate TrustPilot scores from user reviews. Did my best to follow the official description of calculations employed by TrustPilot themselves.

Example

• Download or scrape TrustPilot reviews for a given company/product. I used the following schema review_author_id, review_created_at_utc, review_star
• using TrustPilotCalc you can run through the entire dataset using playBack(filename) function

  julia> Pkg.add("TrustPilotCalc")

• if you are wanting to parse your own dataset e.g. from a db, use getTrustScore(reviews) function and pass in an array of TPReview objects.

using TrustPilotCalc
using Dates, JuliaDB, Lazy

# Import sample TrustPilot data
example_data = loadtable("examples/trustpilot_sample_data.csv"; type_detect_rows = 500)

# Example of a time-series of monthly TrustPilot scores
@as data example_data begin
  unique(map(x-> round(x.review_created_at_utc, Dates.Month), data))
  sort!(data; rev = true)
  pop!(data)
  [print(playBack(example_data; max_yearmon = x).score, " | ", x, "\n") for x in data]
end

Data sourcing

With a TrustPilot business account you can export all your reviews as a flat csv file.

Alternatively, here is an example Javascript function to generate a quick dataset from a TrustPilot reviews page. Just run it in your browser console.

(function(){
function parse_text(text) {
  let output = '';
  try {
    output = text.textContent.trim();
  } catch(e) { ''}
  return output;
};
function process_review(review) {
  data_review_info = JSON.parse(parse_text(review.querySelector('script[data-initial-state=review-info]')));
  data_review_dates = JSON.parse(parse_text(review.querySelector('script[data-initial-state=review-dates]')));

  review_blob = {
    review_id: data_review_info.reviewId,
    review_created_at_utc: data_review_dates.publishedDate,
    review_author_id: data_review_info.consumerId,
    review_username: data_review_info.consumerName,
    review_email: '',
    review_title: parse_text(review.querySelector('.review-content__title a')),
    review_content: parse_text(review.querySelector('.review-content__text')),
    review_star: data_review_info.stars,
    review_source: '',
    reference_id: '',
    company_response: parse_text(review.querySelector('.brand-company-reply__content')),
    review_language: '',
    domain_url: review.getAttribute('data-review-domain-name'),
    webshop_name: data_review_info.businessUnitDisplayName,
    business_unit_id: data_review_info.businessUnitId,
    tags: '',
    company_reply_at_utc: review.querySelector('.brand-company-reply__date time') ? review.querySelector('.brand-company-reply__date time').getAttribute('datetime') : '',
    location_name: parse_text(review.querySelector('.consumer-information__location span')),
    location_id: '',
    reviewer_review_count: Number(parse_text(review.querySelector('.consumer-information__review-count span')).replace(' reviews',''))
  };
  return review_blob;
};

function to_flatformat(review_objs) {
  keys = Object.keys(review_objs[0]);
  rows = Array.prototype.map.call(review_objs, function(obj){return Object.values(obj)});
  blob = Array.prototype.concat([keys], rows);
  return Array.prototype.map.call(blob, function(line){return line.join("','");}).join("'\n'")
};

_reviews = document.querySelectorAll('.review');
_review_objs = Array.prototype.map.call(_reviews, function(review){return process_review(review)});
return to_flatformat(_review_objs);
})();

References

• High level overview of TP's approach to calculating scores: description of TrustPilot's scoring algorithm
• Evan's blog post on bayesian averages:
• Bayesian average scoring blog post
• Another blog post on rating analysis