CoursesTeachersMethod

Coursera says 4.9.
The reviews say something else.

Starisk reads the written reviews — and the votes other learners gave them — then shows what the most-endorsed reviews actually say.

Every rating deserves an asterisk.

Where the headline rating is inflated

Topic
Sequences, Time Series and Prediction
DeepLearning.AI
1.5
3.2
Dog Emotion and Cognition
Duke University
2.0
2.7
Natural Language Processing with Attention Models
DeepLearning.AI
1.7
2.7
Divide and Conquer, Sorting and Searching, and Randomized Algorithms
Stanford Online
2.1
2.7
People & Soft Skills: Essential for Professional Success
IBM
2.1
2.6
R Programming
Johns Hopkins University
1.9
2.6
Statistical Inference
Johns Hopkins University
1.6
2.6
Getting and Cleaning Data
Johns Hopkins University
2.0
2.5
Practical Machine Learning
Johns Hopkins University
2.0
2.5
Collaborate Effectively for Professional Success
IBM
2.1
2.5
Reproducible Research
Johns Hopkins University
2.2
2.4
Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning
DeepLearning.AI
2.5
2.3
Natural Language Processing with Sequence Models
DeepLearning.AI
2.2
2.3
Convolutional Neural Networks in TensorFlow
DeepLearning.AI
2.5
2.2
Android App Components - Intents, Activities, and Broadcast Receivers
Vanderbilt University
2.0
2.2
Mathematics for Machine Learning: PCA
Imperial College London
1.8
2.2
Natural Language Processing in TensorFlow
DeepLearning.AI
2.4
2.2
Operating Systems: Overview, Administration, and Security
IBM
2.4
2.2
Spanish Vocabulary: Meeting People
University of California, Davis
2.4
2.2
Miracles of Human Language: An Introduction to Linguistics
Universiteit Leiden
2.6
2.1
Learn Spanish: Basic Spanish Vocabulary
University of California, Davis
2.4
2.1
Exploratory Data Analysis
Johns Hopkins University
2.6
2.1
Sequence Models
DeepLearning.AI
2.7
2.1
Mathematics for Machine Learning
Imperial College London
2.5
2.1
Introduction to Cybersecurity Tools & Cyberattacks
IBM
2.5
2.1

Showing 1–25 of 316 (426 total)

Page 1 / 13

Sorted by inflation: Coursera's rating minus Starisk's — how many stars the headline sits above what the most-upvoted reviews say. Inflated ≥ 1.5,padded ≥ 0.5, otherwise it holds up. Sort by any column.

Method, honestly

We read the reviews learners voted most helpful — about 20 per course — and weight each one by the square root of its votes: a review with 80 upvotes counts about five times one with 2. Enough to matter, not enough to decide alone. Without that compression a single review with 663 likes would outvote everything else on the page, and on the 29 courses I labelled myself that fit was measurably worse.

Coursera publishes the full rating distribution for every course, so each page shows the sample next to the whole population — including what share of all ratings are 1–2★. That is the comparison worth making: a small upvoted minority, against tens of thousands of five-star ratings.

It measures what the most-endorsed reviews say, not “true quality.” Upvoted reviews often carry a grievance — about access, pricing or certificates — rather than a judgment of the teaching. The disagreement is the point, not the final word. If a course had fewer than 15 readable reviews, its page says so instead of pretending.

Sample: 426 courses and 151 teachers. Scores recomputed from saved review pages; not affiliated with Coursera.

The full method, with the calibration numbers →