In this study, we have developed an Indonesian WordNet through four main phases: synonym set extraction (synset) as the smallest entity of lexical database from a natural language, semantic relation establishment between synsets (hypernym-hyponym and holonym-meronym), gloss extraction for synset collection, and the visual editor creation. The Semi-automatic term refers to the three initial phases which are automatically done using a number of machine learning approaches, while using visual editor to collaboratively complement the results collected from the previous phases. A number of raw data used on synset acquisition, semantic relations and glosses come from Kamus Besar Bahasa Indonesia (Great Dictionary of the Indonesian Language, abbreviated as KBBI) and Tesaurus Bahasa Indonesia (Indonesian Language Thesaurus), large collection of web pages from search engines, Wikipedia, and even Princeton WordNet for mapping purpose. This study shows that the proposed system successfully achieve 37,485 synsets, 24,256 hypernym-hyponym relations, 11,044 holonym-meronym relations and 6,520 gloss synsets. Similar approach is believed to accelerate lexical database development like WordNet for other languages.