Introduction

One of the biggest obstacles for users switching to Linux is choosing the right distribution. With the end of support for Windows 10 on October 14, 2025, many users are looking for alternatives to continue using their devices.
This project aims to help users migrate from Windows 10 to GNU/Linux by analyzing their system specifications, asking relevant questions, and recommending a ranked list of distributions with KDE Plasma based on their preferences.

This post will break down how we approached this problem in SoK25, focusing on two core challenges:

  1. Designing a recommendation logic that ranks distributions based on user preferences.
  2. Detecting hardware specifications and integrating them into the ranking logic.

Original Idea

Step 1: Initial Design for the Questionnaire

When we started the project, we considered including a range of distributions and desktop environments. Over the course of SoK, however, we decided to limit the recommendations to distributions offering KDE Plasma. I will first present the original idea.

To keep things simple, we decided to limit the number of questions we would ask Windows users. Instead of overwhelming users with technical details, we settled on a set of key questions. These cover parameters like:

  • Ease of installation
  • Support for older hardware
  • Quality of documentation
  • Community support
  • Frequency of updates (Rolling or Point)
  • Preferred UI style (Windows-like or Mac-like)

Each response contributes to a “parameter” of user preference vector, which helps in matching the needs of user with the best-fitting distributions.

Step 2: Selecting Initial Distributions

We initially started with a limited set of well-known beginner-friendly distributions:

  • Fedora KDE
  • Kubuntu
  • Lubuntu
  • Linux Mint XFCE
  • Linux Mint Cinnamon

Revised Idea

Over the course of SoK25 we decided to narrow the scope of the chooser app and rename it Win10-2-KDE-Chooser. After receiving some feedback from the community, we decided it made sense to focus on KDE-only solutions for a Season of KDE project, while others can adapt the idea at a later date.
With this new scope, we think the app can be promoted by the KDE community in the context of the ‘End Of 10’ campaign. The new scope includes:

  • The app will target devices running Windows 10.
  • The app will recommend distros with KDE Plasma.
  • The recommended distros for first-time users include Fedora KDE, Kubuntu, and Debian KDE (specifically the live installer, which uses Calamares).

In the rest of this SoK25 post, I will describe the implementation of the original idea in more detail, and this can later be adapted before release. This has been the focus of my work for the first half of SoK25.

Step 3: The Recommendation Logic

The core logic behind ranking distributions is vector-based similarity measurement. Each distribution is represented as a vector, with dimensions corresponding to the parameters defined in our questionnaire.

The approach works as follows:

  1. User Preference Vector: The answers provided by the user form a vector with numerical values assigned to each preference.
  2. Predefined Distribution Vectors: Each distribution has a corresponding vector based on predefined scores.
  3. Similarity Calculation: The similarity between the user vector and each distribution vector is computed using a mathematical function.

TL;DR : We are using dot product and a penalty criteria (multiplying by 0.5) for ranking the distros

Choosing the Right Similarity Function

Initially, we considered cosine similarity, but we found that dot product gave better results.

  • Cosine Similarity measures how closely two vectors align in terms of direction, but it ignores magnitude.
  • Dot Product considers both direction and magnitude, making it a better fit since we care about absolute scores.

Example:

example{width=456 height=164}

According to dot product: A>B>C Dot Product recommends Distro A as the best choice, which makes sense.

But according to cosine similarity: C>B>A

Distro C appears as the best match because it follows the same ratio (even though its scores are much lower). Since we care about absolute quality in preferred criteria rather than just proportional similarity, dot product is the better approach.

Here’s how we implemented the ranking in Python:

    def recommend(self):
        if not self.user_vector or not self.distro_vectors:
            return "No sufficient data to generate recommendations."

        scores = {}
        user_vector_np = np.array(self.user_vector)

        for distro, vector in self.distro_vectors.items():
            distro_vector_np = np.array(vector)
            score = np.dot(user_vector_np, distro_vector_np)

            scores[distro] = score

        sorted_indices = np.argsort(list(scores.values()))[::-1]
        ranked_recommendations = [(list(scores.keys())[i], list(scores.values())[i]) for i in sorted_indices]
        return ranked_recommendations

Step 4: Handling Categorical Questions

While numerical parameters like “Ease of Installation” are easy to quantify, categorical preferences (e.g., “Do you prefer a Windows-like UI?”) are more like binary or ternary preferences and are difficult to score.

To handle this, we introduced a penalty mechanism:

  • If a distribution does not match the user’s categorical preference, a penalty factor (e.g., 0.5) is applied to its score.
  • This ensures that distributions aligning with strong user preferences are ranked higher.

Here’s the modified code which ranks distros with penalties:

  def recommend(self, penalty_factor=0.5):
        if not self.user_vector or not self.distro_vectors:
            return []  # Return an empty list instead of a string

        scores = {}
        user_vector_np = np.array(self.user_vector)
        binary_params = {"updates", "UI_Look"}

        for distro, data in self.distro_vectors.items():
            distro_vector_np = np.array(data["scores"])
            final_score = np.dot(user_vector_np, distro_vector_np)

            for param in binary_params:
                if param in self.user_binary_preferences and param in data["raw_scores"]:
                    if self.user_binary_preferences[param] != data["raw_scores"][param]:
                        final_score *= penalty_factor

            scores[distro] = final_score

        return sorted(scores.items(), key=lambda x: x[1], reverse=True)

Conclusion

This ranking system forms the backbone of the chooser app, helping users find the most suitable distribution based on their needs. While our initial model looks good, we are still refining the parameters (questionnaire), expanding the dataset, and revising the app to fit the new scope

Next steps include:

  • Enhancing hardware detection to factor in compatibility scores. Given the limited time left in SoK25, I will only focus on Nvidia driver detection.
  • Improving penalty logic for better handling of categorical preferences.

This project has been an exciting challenge and I’m looking forward to refining it further. Stay tuned for more updates!

Acknowledgement

Thank you to the Season of KDE 2025 admin and mentorship team, in particular flyingcakes an Aakarsh MJ and Joseph, the KDE e.V., and the incredible KDE community for supporting this project.

Please feel free to contact me here: @drowsywings:matrix.org