Scientists have been using scientific workflow management systems (SWfMS) to support scientific experiments. However, SWfMS expect a modeled workflow to be represented on its workflow language to be executed. The scientist does not have an assistance or guidance to obtain a modeled workflow. Experiment lines, which are a novel approach to deal with these limitations, allow for the abstract representation and systematic composition of experiments. Since there are many scientific workflows already modeled and successfully executed, they can be used to leverage the construction of new abstract representations. These previous experiments can be helpful by identifying scientific workflow clusters that are generated according to similarity criteria. This paper proposes SimiFlow, which is an architecture for similarity-based comparison and clustering to build experiment lines following a bottom-up approach.