The relation extraction task which aims to identify the relationship between a specified pair of words is considered a significant task that can be expanded to be utilized in various ways. Therefore, the automatic relation extraction system is often considered a key to extracting systematic reusable information from sentences, paragraphs, or documents. Instead of achieving state-of-the-art performance, this thesis aims to explore the significance of various input & output representations, and the difference between a linear and a bilinear classifier on relation extraction tasks. For a thorough analysis, I experiment on a diverse group of relation extraction datasets and present a set of ablation studies. Moreover, experiments are compared not only based on their performance but also on the efficiency of resource usage. The analysis illustrates that the systems based on certain input & output representations yield the best performance in general even though introduced systems have less complexity compared to bilinear-classifier-based systems. Moreover, the straightforward systems studied in this thesis show results comparable to state-of-the-art systems (3% difference) in general.