Ideation contests receive thousands of ideas. How to filter the most innovative ones? We propose a novel method involving the crowd to solve this problem. PDF to full paper: https://tinyurl.com/2s379kyf
When Crowds Give You Lemons: Filtering Innovative Ideas using a Diverse-Bag-of-Lemons Strategy
1. When Crowds Give
You Lemons:
Filtering Innovative
Ideas using a
Diverse-Bag-of-
Lemons Strategy
Ioanna Lykourentzou, Faez Ahmed,
Costas Papastathis, Irwyn Sadien,
Konstantinos Papangelis
3. How to filter the
most innovative
ideas involving
the crowd?
• Example: Redesigning the
Bus Experience
• 36 ideas
• Symbols indicate the broad
theme of each idea
6. Snowball effect
• Ideas which
received the first
initial upvotes are
more likely to be
seen by
subsequent users
7. Problems with majority voting
• Prone to quickly locking into a fairly static ranking
due to positive feedback loops
• Crowds are less effective in distinguishing mediocre
from excellent ideas
• Prone to quickly locking into a fairly static ranking
due to positive feedback loops
• Crowds are less effective in distinguishing mediocre
from excellent ideas
8. Inspiration:
Bag of
lemons
• Bag of Lemons: Ask
voters to identify the
worst rather than the
best ideas using a
multi-voting
approach (Klein and
Garcia (2015))
9. DBLemons: Diverse
Bag of Lemons
We propose a crowd-voting method, which
overcomes these problems by using Bag of Lemons
voting and exposes voters to a wider idea spectrum,
thanks to a dynamic diversity-based ranking
system
Algorithm implemented in three steps:
1. Identify clusters to which each idea belongs
2. Rank order ideas using our algorithm
3. Ask crowd worker to vote using lemons
10. DBLemons Trade-off
Function
• Utility function is submodular, hence a greedy
algorithm is theoretically guaranteed to provide the
best possible polynomial-time approximation to the
optimal solution.
19. Women Safety
Ideation Contest
• The dataset we worked with
• How might we make low-income
urban areas safer and more
empowering for women and girls?
20. Process
• We created 52 idea set summaries of existing
OpenIDEO ideas.
• Each summary was reviewed sequentially by 3
reviewers. Each idea evaluated by 30 crowd
workers on 5-point Likert scale.
• Top 30% selected as golden set
• Comparison between majority voting, Bag of
Lemons (BOL), and DBLemons approaches
22. Results
• DBLemons is more
accurate, less time-
consuming, and
reduces the idea
space in half while
still retaining 94% of
the top ideas.
Performance comparison of the three ranking
strategies. The dashed line shows the
golden set cardinality cut-off
23. Results
• Progressive ranking per strategy, descending quality order.
All strategies improve with the number of voters, but
DBLemons does so faster
25. Results
• People made multiple
hops to compare
ideas with each other
but took less time
overall
26. Contributions
We show that:
• Majority voting is more time consuming and less
accurate than the other two alternatives
• Balancing idea quality and idea representativeness
improves the filtering of high-quality ideas
• Bag of Lemons with dynamic ranking has similar
filtering efficiency as Majority voting, but finds good
ideas faster. This result confirms the promise of the
technique proposed by Klein and Garcia on the
dynamic setting
27. Future work
• The methods developed have wide
ranging impact on large- and small-
scale ideation contests.
• Few directions of future research are:
• Systematically study how to select
number of lemons
• Study scalability challenges for very large
idea datasets