| Intercultural Collaboration Environment (ICE) is a joint research project of universities, research institutes, and research societies in Asia. The objective of ICE is to support intercultural and multilingual collaborations using machine translation technologies (MT). ICE started from 2002 has attracted many universities, research institutes, and research societies in Asia.The translation tools such as TransBBS, AnnoChat, which were developed by ICE group, are able to translate messages among Chinese, English, Japanese, and Korean, the outcome of the test suggested that the comprehensibility should be improved.To improve the mutual understanding of users using their native languages in Asia, an ontology-based ICE framework was proposed in which ontology, agent and data mining techniques were integrated. A prototype called OBICES (Ontology-Based ICE System) was implemented based on the proposed framework. With the support of this framework, the translation results of AnnoChat are easier to be understood. The work of this thesis is focused on the text mining module.There are two main tasks in the text mining module of OBICES: Firstly, mining the semantics of the on-line chat text instantly and assisting agents to make the semantic choice, Secondly, mining historic text semantics and theme in the chat database. The knowledge mined by the two tasks should be saved in the chat knowledge database and it is also used to revise and refine the Domain Ontology which is used to provide background knowledge for the users.This research improved the Term Frequency Inverse Document Frequency in the text mining process according to the character of the online chat text, and proposed atext mining method to define the dynamic window size.According to the characteristics of the online chat text, a dynamic window size text mining method was proposed, which is based on the Term Frequency Inverse Document Frequency algorithm. The method has been verified by experiment. The experimental results show that it can improve the effect and quality of the text mining.By comparing the results of the comprehensibility of keywords, sentences and theme with and without the support of OBICES, it is identified that OBICES can improve the comprehensibility of the translation results of AnnoChat. |
