| Summary: | The aim of this paper was to develop a system for identification of biomedical entities, such as protein and gene names, from a corpora of Medline abstracts. Another aim was to manage to extract the most relevant terms from the set of identified biomedical terms and make them readily presentable for an end-user. The developed prototype, named iMasterThesis, uses a dictionary-based approach to the problem. A dictionary, consisting of 21K gene names and 425K protein names, was constructed in an automatic fashion. With the realization of the protein name dictionary as a multi-level tree structure of hash tables, the approach tries to facilitate a more flexible and relaxed matching scheme than previous approaches. The system was evaluated against a golden standard consisting of 101 expert-annotated Medline abstracts. It is capable of identifying protein and gene names from these abstracts with a 10% recall and 14% precision. It seems clear that for further improvements of the obtained results, the quality of the dictionary needs to be increased, possibly through manual inspection by domain experts. A graphical user interface, presenting an end-user with the most relevant terms identified, has been developed as well.
|
|---|
