Feature extraction and clustering-based retrieval for mathematical formulas

Mathematical formulas or expressions are essential for presenting scientific knowledge in many research documents in academic areas such as physics and mathematics. Searching for related mathematical formulas is an important but challenging problem as formulas contain both structural and semantic information. Such information is hidden inside the mathematical expressions of the formulas. To support effective formula search, it is necessary to extract the structural and semantic features from the mathematical presentation of the formulas faithfully. In this paper, we propose an effective approach for formula feature extraction. To evaluate the proposed approach, the extracted features are tested with three popular clustering algorithms, namely K-means, Self Organizing Map (SOM), and Agglomerative Hierarchical Clustering (AHC), for formula retrieval. The performance of the clustering-based retrieval is measured based on a dataset of 881 formulas and promising results have been achieved.