Kansei Data Mining

Product design in the era of sensitivity

In recent years, any product can easily be acquired on the Internet, enabling us to make choices in terms of-performance, functionality and price. While this is good for consumers, it challenges developers to find ways of continuously improving product affect.

For example, among the many notebook PCs or flat screen TVs at an electric appliance shop, whose functions and prices are approximately equal, how do you choose the one that’s best for you? Because ordinary users often fail to recognize slight functional differences between products, they may finally select one whose appearance pleases them more.

The car—one of Japan’s most important industrial products, provides another good example. In buying a new car, we may consider its size, price, function and fuel economy. To decide which car is best for you as a driver, you are likely to take repeated test drives, but many will ultimately judge a car by its appearance.

Investigating user sensitivity

What type of design attracts users?

This question is crucial for product development. One general method for investigating user preferences is the questionnaire survey. We collect product samples appearing in the market and use questionnaires to explore the impressions of large samples of users.

We then examine the collected data, using factor analysis or principal component analysis to discover why users find a particular design more attractive than others.

Here’s an example.

In this figure, the exterior design of a car is investigated using a general Kansei analysis. The car’s exterior design can be described in terms of various component parts, including shape and size of headlights and radiator grille, size of wheel and thickness and angle of pillar.

These components provide visual stimulation for users, leaving them with impressions of various kinds.

In this analysis, user impressions are quantified by means of a questionnaire survey, and the system learns the relationship between design factors and user impressions. The system can then estimate the likely change in user impression when a design factor is changed.

This estimation is not straightforward because user impressions are really complicated, and different individuals may form different impressions of the same car. In addition, a huge number of design factors go into shaping the appearance of a car, making it almost impossible to identify all the factors that influence user impressions.

Layering impressions

To solve the problem described above, we developed the following hypothesis about user preferences:

Although user product preferences are variable, there may be individual differences regarding each element that can be evaluated by an objective standard.
Look at the lower figure. Now, I would like to ask you a question.

Do you think the front lights of the car are round?

Strictly speaking, the correct answer is that they are not round but oval. But if you are asked such a question, you would probably answer “Yes, they are round.”

Here’s the next question.

Do you think a car with round headlights is pretty?

This may be difficult to answer if you know only about the headlights of the car. So, would you expect most people to feel that a car with round headlights generally looks prettier than one with sharper, squarer headlights?

And finally, one more question.

Do you like a car that looks pretty?

Those three questions focused on round, pretty and like, respectively. Those three words are evaluated by human sensitivity. The answers would differ on the basis of individual preferences.

This does not mean that roundness is measured accurately, and the answer depends on the sensitivity of individual user—in other words, it depends on the user’s Kansei.

As mentioned above, while evaluation of roundness is likely to differ between users, preference is different again. On that basis, we believe there are various levels of Kansei, some of which are objective and others which are very subjective. This suggests that these evaluation words are organized as a hierarchical structure.

On that basis, before checking the relation between design elements and user preferences, we investigated “partial impressions” that would be influenced by partial components or design elements.

We developed a method to visualize relations between partial and global evaluation of attributes by a technique known as the fuzzy decision tree.

Rather than just identifying which product users prefer, this method enables us to establish why they prefer it.

We are now developing a visualization technique for data analysis and a software user interface to enable all designers and developers to more easily use this method of investigation.

References

M.Tokumaru, N.Muranaka, “Product-Impression Analysis Using Fuzzy C4.5 Decision Tree”, Journal of Advanced Computational Intelligence and Intelligent Informatics, Vol.13, No.6, pp.731-737, 2009-11.

T.Matsuo, H.Katayama, M.Tokumaru, N.Muranaka, “Sensitivity Information Analysis of Running Shoes Using Fuzzy Decision Tree and Visualization of Analytical Results”, Joint 5th International Conference on Soft Computing and Intelligent Systems and 11th International Symposium on Advanced Intelligent Systems (SCIS&ISIS2010), pp.156-160, 2010-12 (Okayama, Japan).

M.Tokumaru, N.Muranaka, “Kansei Impression Analysis Using Fuzzy Decision Tree”, Proc. of International Conference on Kansei Engineering and Emotion Research 2010, pp.1333-1342, 2010-03 (Paris, France).