Poznan Contribution to TREC-PM 2019

This paper describes Poznań contribution to the Precision Medicine track of the TREC 2019. In this submission we present several novelties. We cover the motivation for the hand-picked values of the weights assigned to the expanded query terms. We propose a result fusion method – slightly modified version of Borda Count algorithm. Additionally we use a learning to rank environment, we analyze the effectiveness of such an approach in combination with our other methods and analyze the achieved results. We also discuss our dedicated document processing methods. We achieve an improvement of up to 0.02 (infNDCG measure) over the baseline for Clinical Trials with our proposed methods, however the evaluation value of our baseline is much lower than the median of all contributions. The reverse effect happens in the Scientific Abstracts task, the baseline we propose is much stronger than the median, but the default setting of learning to rank proposition lowers the overall evaluation score.