Abstarct: Today, people's research journey begins by entering information needs in an information system such as web search engines. If a person searches for a mathematical formula to find a solution to a problem or to get information about it, his information need is in the field of mathematical information retrieval. Mathematical information retrieval involves searching for specific mathematical concepts, objects, or results, usually represented by mathematical formulas. Although mathematical formulas play an important role in the transmission and dissemination of scientific information, the retrieval of mathematical information is not well supported in most information systems. Formulas are usually represented in their processable form, as text or complex tree structures that describe the appearance or syntax of the mathematical formula. The complexity, hierarchy, and semantic relationships between symbols in mathematical formulas make the models developed for text retrieval unsuitable for mathematical information retrieval. In this research, a mathematical formulas recovery model called TAGLTR is proposed, which uses support vector machine to learn the ranking of mathematical formulas. To improve the performance, this model combines several similarity features baxsed on tree matching and one similarity feature baxsed on graph embedding of mathematical formulas, which have a good performance in modeling the similarity between formulas. The recovery results obtained from the proposed model show an increase in the accuracy and recall criteria compared to the reference models.