Along with the times, the number of scientific papers is increasing. Demands for information retrievals in scientific papers have also increased. Regarding experimental scientific papers, researchers have difficulty in searching for information on experimental scientific papers because information retrieval engines have limitations in the search process due to text mining-based feature extraction of the entire text, while experimental types of scientific paper have specific contents, which should have a different treatment in feature extraction. In this paper, we propose a new system for information retrieval on experimental scientific papers. This system consists of 4 main functions: (1) Specific content-based feature extraction, (2) Classification model, (3) Context-based subspace selection, and (4) Context-dependent similarity measurement. In feature extraction, our system extracts feature category in experimental scientific papers with specific content-based features, which are data, problem, method and result. For the classification model, we use several classification algorithms to classify the specific content features of query papers to supervised document aggregation. In Context-based Subspace Selection, the system carries out dimension reduction with context-based subspace selection selected by the user. To obtain final search result, we make a similarity measurement in context by constructing context-dependent dataset metric to the papers. To perform the applicability of our proposed system, we tested 77 papers in the dataset with the Leave-One-Out validation model with several classification algorithm (Nearest Neighbor, Naive Bayes, Support Vector Machine and Decision Tree) and on average performed 66.65% precision rate and accuracy of 76,18% precision rate. We also made the experiment on the similarity measurement by giving the paper query and the desired content (data, result, method, and problem) as a context given by the user. In the similarity measurement experiment, our proposed system performed 79.17% accuracy rate.