Nowadays, ontology-based recommendation systems are commonplace among the
recommendation methods for alternative products. These ontologies can be enriched by the
usage of knowledge extraction methods such as Formal Concept Analysis. However, online
datasets (i.e., they are being infinitely generated in real time) require certain techniques to
be put in place to maintain a bounded-sized knowledge-base. Consequently, some knowledge
will be lost in the process. In this article, we compare two algorithms to regain a part of the
lost knowledge by merging conceptual structures (thus, enriching the final knowledge-base), on
a real dataset, and present insights based on the comparison.