Presenting a set of similar or diverse ideas during the idea generation process leads ideators to come-up with more creative and diverse ideas. However, to better assess the similarity between the ideas, we designed two matching systems, namely I-Match and OntoIdea. In the context of the idea generation process, each idea is represented by a set of instances from DBpedia describing the main concepts of the idea. Then, the developed matching systems are applied to compute the similarity between a set of instances that represent the ideas. The purpose of our participation at OAEI is to evaluate our designed instance matching algorithm in order to apply it to assess the similarity between ideas. The results obtained for the first participation of I-Match and OntoIdea systems at OAEI 2017, on different instance matching tracks are so far quite promising.
The idea generation process is the key part of innovation. This process aims to generate ideas to solve problems and challenges. A promising approach for supporting such process is the ”brainstorming method” [3]. This method seeks to increase the number of ideas based on ideas of collaborating individuals while restricting criticism.
In addition to leveraging the crowd [10], prior work has shown that generating ideas
that are both creative and diverse can be greatly enhanced through presenting
inspirational examples [6]. However, a major issue is ”how to find inspiring ideas from
hundreds” [9]. To overcome this challenge, research has shown three ways of selecting a set
of inspiring examples systematically [4, 5]: (
Our work is in line with approaches that assess the diversity (i.e. low similarity rating) of inspiring examples automatically [8]. However, assessing similarity between ideas is challenging due to the form of the ideas, i.e. the ideas are described in a short unstructured text.
To solve this problem, we propose another strategy from our prior work proposed
in [2]. This strategy consists of two main parts: (
In this paper, we focus on the matching part of the proposed solution by describing our two instance matching systems I-Match and OntoIdea. The designed systems implement an enhancing algorithm that we proposed in our previous work [1]. The proposed algorithm extracts first all information about the two instances to be matched and normalizes them using NLP. Then, it applies edit distance as a matcher to calculate the similarities between the normalized information. Finally, the approach selects the equivalent instances based on the maximum of shared information between the two instances.
We summarize the algorithm of our developed systems to provide a general idea of the proposed solution. It consists of the following successive phases: 2.1
The system extracts from each individual Ii P1 m1; P2 m2,... a set of information m1,
m2, ... using different properties P1, P2, .... Then, NLP techniques are applied to
normalize these information. In particular, three pre-processing steps are performed: (
In this step, the system calculates the similarities between the normalized informations using edit distance as string matcher. Our system selects the maximum similarity values calculated between different informations by edit distance. If two informations are the same (based on maximum similarity values) the counter is incremented to 1, etc.
Finally, we apply a filter on maximum counter values in order to select the correspondences which mean that the selected correspondences (equivalent individuals) are those who share maximum informations. 3
The I-Match and Ontoidea systems participated only for instance matching tracks of OAEI 2017 evaluation campaign. For the results Please refer to the following website: http://oaei.ontologymatching.org/2017/results/index.html.
In this paper, we have introduced I-Match and OntoIdea, two systems specially designed to compute similarity between instances. The proposed algorithm is useful, especially when the instances contain terminological information. The developed systems provide a quite promising results, thus, we will be applied in the context of the idea generation process to asses similarity between ideas.
As future perspective, we attempt to apply enhance our instance matching algorithm especially for DORUMUS track.
Acknowledgements The authors acknowledge the financial support of the Federal Ministry of Education and Research of Germany in the framework of Ideas to Market (project number 03IO1617).