Despite the progress in bioNLP domain, the majority of text mining methods/techniques are not used in the real curation tasks. Partly, it can be explained by the lack of precision and opacity of the provided results. Furthermore, the functionality and adequacy of systems, designed for the sake of curation are also important. In this thesis, the original methods used in the context of biocuration assistance are first explored and developed. More specifically, the focus of the thesis is on how text-mining systems are used in the processes of curation. Finally, the results of an assisted curation are compared to the ones of manual curation. The achieved improvement in assisted curation accuracy is +9.3% in average. Although the evaluation of the curation tasks is based on a small sample of participants and benchmark, the achieved results suggest that the improvement of the curation quality is possible. Moreover, it is also dependent on the expert skills.