Nanoblog processing

点击翻译按钮获取中文摘要

Natural Language Processing (NLP) has traditionally been done with long documents, and tasks with such documents achieve very high performances and many are even considered "solved". The popularity of SMS messaging and the creation of social networks led to the popularization of short text messages or "microblogs", and many previously-solved NLP tasks became open challenges again. New difficulties stemmed from the short length of the messages, the limited contextual information available, the use of unconventional language, and the socially-linked and dynamic nature of the conversations. Nevertheless, after a decade of research work, many of these challenges have been overcome and microblog-processing systems routinely achieve performances comparable to those in traditional NLP. And thanks to these advances, microblogs have become a valuable resource in other research areas as well. The next frontier in NLP will come from increasingly-popular in-game chat messages found in online games and gaming communities. These messages amplify the problems of microblogs: they are even shorter, consisting of sometimes only a single word or acronym and completely devoid of in-message context, they use language tailored to the specific game they are linked to and that is gibberish outside of that game, and only make sense as live conversations during a game. Due to their extremely short nature, we call these conversations "nanoblogs". The long-term objective of this research proposal is to pioneer the area of nanoblog NLP and ensure Canada's leadership in this emerging research area. To achieve this, we have defined three short-term research objectives to accomplish over the next five years. Objective 1: We will design, construct, and label a real-world nanoblog corpus. Objective 2: We will create nanoblog context modelling, normalization, and language detection algorithms. Objective 3: We will create nanoblog cyber-bullying filter algorithms.