Tableau dashboard · 2026 · by Luiz Trajano
Trending YouTube Videos Dashboard
Which video categories trend, how do they split across countries, and what stands out in the US? Answered every week, without an analyst in the loop.
Tableau · Tableau Public · Pandas
Origin: TripleTen Data Analyst track (sprint 12). Brief and data provided by the course; analysis, code and conclusions are mine.
The question
Every week, the new planners at an ad agency asked the same three things: which video categories were trending last week, how they were spread across regions, and which categories were especially popular in the United States. The brief was to stop answering by hand and build a dashboard they could open themselves, at least once a day, with a date filter and a country filter that drive every chart at once.
The layout came from a sketch agreed with the managers: title and description top left, filters top right, then a 2 × 2 grid. Trending history in absolute numbers as a stacked area, the same history as a share of the total, a pie by country, and a highlight table with countries in the columns and categories in the rows.
The data
One aggregated table, trending_by_time: 12,343 rows, one per country, day and category, from
14 November 2017 to 14 June 2018. Five countries (France, India, Japan, Russia, United States) and
18 categories. No nulls, a unique key, and no duplicates on country, day and category. Every
timestamp sits at midnight, so the real grain is daily.
Three properties of this table change how the dashboard has to be read, and I found them before drawing anything:
videos_countcounts video-days, not distinct videos. A video trending for five days enters five times, so the 339,990 total is a daily sum.- Japan starts on 7 February 2018. It has 122 days of data against 205 for every other country.
- Eight days have no data at all: 10 and 11 January and 8 to 13 April 2018.
Method, and why this method
I checked every number in pandas first (validacao.ipynb) and treated that notebook as the answer
key: the dashboard was only done when it matched it.
In Tableau, the date field went in as a real date with a continuous range filter, not as text. The
CSV carries six decimal places on the seconds, and Tableau tends to import that as a string. Because
YYYY-MM-DD sorts the same way as text or as a date, every chart still looks right and only the
filter gives it away.
The percentage history uses a percent-of-total table calculation computed along the category dimension, so each day adds up to 100%. The test is visual and strict: the top of the area has to be a flat line at 100%. The date and country filters are applied to every worksheet that uses the data source, so the four views move together.
Two questions needed a decision before any answer. "Last week" has no reference date in the brief, so I fixed it as the last week in the data, 8 to 14 June 2018. And comparing the US with the other countries in raw counts would make the biggest market look bigger at everything, so I compared the share of each category within each group, in percentage points.
Findings
Entertainment leads with 27.9% of all trending video-days and is first in four of the five countries. Russia is the exception, where People & Blogs leads with 25.0%. In the week of 8 to 14 June, Entertainment rises to 30.6% and Music (12.1%) moves ahead of People & Blogs (11.1%), so the weekly cut reorders the ranking.
Over the whole period, the United States, France, Russia and India sit between 21.6% and 23.8% of the total and Japan shows 10.8%. That number is the calendar talking. Restricted to the window where every country has data, Japan rises to 17.0% and the other four move down, as the chart above shows. Read at face value, the pie would say that Japan watches less trending content.
In the US, Music (+7.6 pp), Howto & Style (+5.7 pp) and Science & Technology (+3.8 pp) weigh more than in the other countries; People & Blogs (−7.3 pp), News & Politics (−5.4 pp) and Entertainment (−4.8 pp) weigh less.
The decision it enables
The planners get their three answers without asking anyone: pick the dates, pick the countries, and the four views agree with each other. The notes on the dashboard itself carry the three caveats, so the numbers cannot be read as more than they are.
For comparisons between regions, start the date filter on 7 February 2018, or the pie reports a difference in collection window as a difference in audience. The table has no view counts, so nothing here measures audience size: it measures how often a category shows up on the trending page. The project was approved by the TripleTen reviewer on the first submission.