0
Research Papers: Design Theory and Methodology

Machine Learning Algorithms for Recommending Design Methods

[+] Author and Article Information
Mark Fuge

Department of Mechanical Engineering,
Berkeley Institute of Design,
University of California,
Berkeley, CA 94709
e-mail: mark.fuge@berkeley.edu

Bud Peters

Department of Mathematics,
Berkeley Institute of Design,
University of California,
Berkeley, CA 94709
e-mail: dbpeters@berkeley.edu

Alice Agogino

Department of Mechanical Engineering,
Berkeley Institute of Design,
University of California,
Berkeley, CA 94709
e-mail: agogino@berkeley.edu

1Corresponding author.

Contributed by the Design Theory and Methodology Committee of ASME for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received January 9, 2014; final manuscript received July 21, 2014; published online August 18, 2014. Assoc. Editor: Irem Y. Tumer.

J. Mech. Des 136(10), 101103 (Aug 18, 2014) (8 pages) Paper No: MD-14-1013; doi: 10.1115/1.4028102 History: Received January 09, 2014; Revised July 21, 2014

Every year design practitioners and researchers develop new methods for understanding users and solving problems. This increasingly large collection of methods causes a problem for novice designers: How does one choose which design methods to use for a given problem? Experienced designers can provide case studies that document which methods they used, but studying these cases to infer appropriate methods for a novel problem is inefficient. This research addresses that issue by applying techniques from content-based and collaborative filtering to automatically recommend design methods, given a particular problem. Specifically, we demonstrate the quality with which different algorithms recommend 39 design methods out of an 800+ case study dataset. We find that knowing which methods occur frequently together allows one to recommend design methods more effectively than just using the text of the problem description itself. Furthermore, we demonstrate that automatically grouping frequently co-occurring methods using spectral clustering replicates human-provided groupings to 92% accuracy. By leveraging existing case studies, recommendation algorithms can help novice designers efficiently navigate the increasing array of design methods, leading to more effective product design.

FIGURES IN THIS ARTICLE
<>
Copyright © 2014 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 3

We performed spectral clustering on the 39 × 39 method covariance matrix revealing groups of methods that covary together. Lighter tones represent low covariance, while darker tones represent high covariance. The different hues denote different clusters, with a dark gray box around each cluster of methods. The clusters found by spectral clustering accurately reflect the expert-given categories used by IDEO in their HCD Toolkit; from left to right, the boxes on the diagonal correspond to “Deliver,” “Hear,” and “Create” methods, respectively.

Grahic Jump Location
Fig. 2

Each case page on HCD Connect contains a textual description about the design problem (1), as well as contextual labels such as location (2), focus area (3), and user occupation (4)

Grahic Jump Location
Fig. 1

HCD Connect users use different methods with different frequencies. Error bars represent 95% confidence bounds around the frequency estimates, calculated using bootstrap resampling. The gray line represents the average method frequency (≈14%).

Grahic Jump Location
Fig. 4

Precision plotted as a function of recall. The higher the AUC, the better the algorithm's performance.

Grahic Jump Location
Fig. 5

The area under the precision–recall curve (AUC) across the models. The error bars represent the 95% empirical confidence bounds about the median AUC for each method, calculated using bootstrap resampling. The hybrid and collaborative filtering models perform substantially better than the popularity baseline. The Random Forest classifier produces a detectable, but small, improvement over the popularity baseline.

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In