Deep Analysis of
1.5 Million Reviews of
Mattress Manufacturers

single blog background
 author`s image

Oleg Boyko

[  ] Jan 2022 – Oct 2022

[  ] 1,498,482 reviews, 20 companies, 32 sources

[  ] Data Scraping, Public API, Python

Key Findnings

1. 3 out of 20 companies started booming after the beginning of COVID-19. 4. Most customers post reviews on Mondays.
2. Customers posting 5-star reviews are more likely to write about personal feelings. 5. Negative feedback is often associated with poor customer service.
3. Customers usually give 1-2 or 4-5 stars. 3 stars are rarely given.

The purpose of the study was to find out what important insights the public data can provide with a help of web scraping using the example of deep analytics of mattress manufacturers product reviews.

For this study, we selected 20 mattress manufacturers in Europe. As a data source, we took reviews from the most popular review aggregators, and have been monitoring them for 9 months. During this time we’ve managed to analyze about 1.5 million unique reviews (1,498,482 to be exact).

We did not analyze the product prices and did not try to find the best manufacturer. We only analyzed the digital shelf of these companies on the Internet.


To get the reviews (text and metadata) we have used 3 categories of sources:

  • Product pages on the Amazon marketplace in various regions
  • Trustpilot Review Aggregator
  • Own marketplaces (online stores) of selected manufacturers.

Data extraction and processing have been done with custom Python scripts that organized and performed all the necessary stages of the ETL pipeline.


    1. Using Google Trends and multiple public listings, we’ve selected the top 20 brands that produce mattresses and other sleep accessories in Europe and North America:
      – Emma – Purple
      – Tuft & Needle – Bett1
      – SimbaSleep – Tempur
      – Casper – Tediber
      – Endy – Nectar
      – Ghostbed – Leesa
      – Eve – Novosbed
      – Logan & Cove – OTTY
      – Silk & Snow – Brunswick
      – Bodyguard – Recore
      – Polysleep – Hamuq
    2. For each manufacturer, we searched for the maximum number of relevant links to marketplaces or individual products and pages that contain product reviews.

      a. On Amazon, regional versions were used for the US, Canada, UK, France, Germany, Poland, Italy, Spain, the Netherlands, Mexico, India, Singapore, Japan, China
      b. Proprietary marketplaces were hosted in the domain zones of the following countries (in addition to those mentioned for Amazon): Australia, Portugal, Belgium, Sweden, Denmark, Austria, Ireland, New Zealand, Taiwan, Philippines
    3. For each site/shop/product, all publicly available reviews have been collected along with the reviews metadata, regardless of their publication date.
    4. Reviews were checked for duplicates. In the case of the full content match for 5 critical data fields (author\title\review body\rating\brand), the review was filtered as a duplicate of an existing review. This made it possible to avoid distortion of statistics when analyzing the reviews on different sources (Trustpilot and own marketplaces).
    5. 1,360,680 unique reviews have been received from 2016 to September 2022. The collected reviews have gone through all stages of content cleaning and deduplication so that we could perform a direct analysis, build statistics, and test hypotheses.


  1. Does the number of reviews from different companies grow linearly over time?
  2. Is there any correlation between the number of reviews and seasonality?
  3. Do customers leave more reviews on any specific day of the week?
  4. What do customers most often like?
  5. What do customers most often dislike?
  6. Does the growth in the number of reviews correlate with the growth of the company (number of employees, turnover)?
  7. What is the industry average ratio of good to neutral to bad reviews?
  8. Are the reviews duplicated across platforms?

Assumption 1: (Does the number of reviews from different companies grow linearly over time?)

The study tells us that new reviews are added linearly over time. However, in the chart, we can see that the dynamics of growth in the number of reviews have significantly increased for 3 companies after the COVID-19 started.

Assuming that the number of new reviews is correlated with new sales, we do not see a negative impact of COVID-19 on the industry. We see positive dynamics for 3 out of 20 companies.

Assumption 2: (Is there any correlation between the number of reviews and seasonality?)

Looking at all available reviews over time we can see a slight seasonality impact. Dynamics change around 10%.

An increase in dynamics can be observed in January and April, and a decrease in November and June.

Assumption 3: (Do customers leave more reviews on any specific day of the week?)

Customers are 50% more likely to leave reviews on Mondays than on the weekends. Perhaps on the weekends customers are testing a new product (mattress) and on Monday they are ready to share their opinion on it.

According to the chart, it’s best to request feedback on the purchased product from customers at the beginning of the week, it can increase the chances of receiving a review.

Assumption 4: (What do customers most often like?)

We used sets of 3 words to determine the topics that buyers like the most. For a more visual result, we used lemmatization, which is ​​a text normalization technique that switches any kind of a word to its base root mode.

We’ve concluded that most often people describe the personal feelings they have after using the product, or the level of service.

The top topics in 5-star reviews were:

  • good night sleep
  • memory foam
  • sleep long time
  • etc.

Assumption 5: (What do customers most often dislike?)

An analysis of negative reviews allows us to conclude that most often buyers do not like customer service.

We also see that sometimes people complain of back pain. Which may mean that the consultant failed to offer a suitable product.

Assumption 6: (Does the growth in the number of reviews correlate with the growth of the company (headcount*, turnover)?)

*Based on LinkedIn data on the number of users who indicated the company as a current place of work.

Comparing the leaders in terms of growth in the number of reviews with their employees’ data, we did not find a direct correlation between these indicators. In general, companies maintain fairly systematic growth over a long period of time. The high rate of growth in reviews, like that of Bett1, doesn’t correlate with the changes in their headcount.

Also, the headcount data makes it possible to mark the period of the 4th quarter of 2021 – the 1st quarter of 2022, during which at least 4 companies faced headcount drops with a gradual growth dynamics recovery in the 2nd quarter of 2022 but at a far slower pace.

Assumption 7: (What is the industry average ratio of good to neutral to bad reviews?)

If we look at all the industry reviews, we can see the following distribution:

  • 89% – 4-5 star reviews,
  • 6.4% – 1-2 stars, and only
  • 4.6% – 3 stars.

Based on statistics, we conclude that buyers prefer positive or negative reviews over neutral ones.

Although, for some companies, the values differ from the industry average. For example, the number of 3 and 4-star reviews for Purple is about the same.

Assumption 8: (If all the reviews made in different languages are unique?)

We’ve conducted additional analysis of the reviews in languages ​​other than English by performing machine translations and analyzing the content for similarity. According to the results, at least one brand uses translated reviews for other locations of their stores. The main advantage of this approach is that the company offers content in a native language for the target audience. However, it can also partly undermine the credibility of the manufacturer due to the fact that translated reviews mostly have unnatural vocabulary because of machine translation. With a content similarity of 85%, about 30% of non-English reviews can be considered non-original.


The results of the study are not aimed to define the best brand in the industry or compile a top list. The purpose was to display the dependencies and changes of specific industry-related parameters in dynamics.

The results prove that the deep analysis of the reviews and their numerical indicators (such as rating distribution, number of reviews, and peaks of reviews publication within a week/month/year) allows to track the dynamics of changes and track the impact of the internal and external factors on the success of the brand on the market. However, it’s worth keeping in mind that a very important factor hasn’t been considered in this study – the dynamics of price changes.

Adding the price as a parameter could help find correlations between the reviews metrics and the success of sales, discounts, bundle offers, the relevance of outdated products support, and how all this impacts the revenue of the company.

With a help of web scraping, the business can get valuable insights and start understanding the market and the competition better.

Looking for a data-driven solution for your retail business?

Embrace digital opportunities for retail and e-commerce.

Contact Us