Nuo Ma -Reflection 1



In this paper, the author address the problem of analogy searching, which is effective when using one domain knowledge to solve problem in another domain, but this is difficult. The author designed 2 experiments, a set of 5 problems was defined and corresponding schemas was provided, then MTurks users was asked to find analogies on, which provide a large dataset of similar idea inspirations. Participants were randomly assigned to 4 conditions: use schemas to avoid introducing unfamiliar terminology for the problem, use surface features to test whether providing features as an intermediate representation would help participants to search for useful and relevant ideas,  find inspiration given no further constraints, and finally solve problems without any intermediate steps to the repository. The results was evaluated by 2 authors of this paper from 3 perspectives, practicality, usefulness and novelty. In conclusion, the author did experiment on searching for analogical ideas from via crowdsourcing and evaluate their qualities. The results shows that, first using schemas could assist in finding more far analogies that did not share surface features with the source problem. The second finding is that analogical ideas previously found can be used to generate better solutions for the source problems.


The methodology in this paper is definitely very interesting to me and has its constraints. The author stated that users are more likely to return product ideas analogous to a problem with given schema than surface features or simply looking for inspiring examples. One interesting finding is that when using, users tend to use images and texts to find inspirations instead of using search bars. What would take to generalize this idea from a small lab study to real world problem solving? Although this is inconsistent with the intention of this research paper, but I would like to make the analogy to usual search or brainstorming process that, given schema (a detailed description of the problem), using a repository of ideas (instead of quirky, can be Google or others) can improve the yield of such process. Coming back to this paper, the constraints are: not all problems have properly described schemas. Also, quirky is a relatively small source of repository (although by itself is huge enough), if to be generalized to ordinary analogy generation scenarios. Another interesting thing is that, when users were asked to use search bars, people type in the keywords of the schema and their synonyms. Although this is not discussed in this paper, but this phenomena is by itself a very interesting research question. The keywords used should be conspicuous enough to tell the main differences, and for the same problem but different users, how would people pick the keywords they use? Would this generate in-category-variance and how much the variance can be?