面向信息检索的自适应中文分词系统

新词的识别和歧义的消解是影响信息检索系统准确度的重要因素.提出了一种基于统计模型的、面向信息检索的自适应中文分词算法.基于此算法,设计和实现了一个全新的分词系统BUAASEISEG.它能够识别任意领域的各类新词,也能进行歧义消解和切分任意合理长度的词.它采用迭代式二元切分方法,对目标文档进行在线词频统计,使用离线词频词典或搜索引擎的倒排索引,筛选候选词并进行歧义消解.在统计模型的基础上,采用姓氏列表、量词表以及停词列表进行后处理,进一步提高了准确度.通过与著名的ICTCLAS分词系统针对新闻和论文进行对比评测,表明BUAASEISEG在新词识别和歧义消解方面有明显的优势.

Information Retrieval Oriented Adaptive Chinese Word Segmentation System

New words recognition and ambiguity resolving have vital effect on information retrieval precision. This paper presents a statistical model based algorithm for adaptive Chinese word segmentation. Then, a new word segmentation system called BUAASEISEG is designed and implemented using this algorithm. BUAASEISEG can recognize new words in various domains and do disambiguation and segment words with arbitrary length. It uses an iterative bigram method to do word segmentation. Through online statistical analysis on target article and using the offline words frequencies dictionary or the inverted index of the search engine, the candidate words selection and disambiguation are done. On the basis of the statistical methods, post-process using stopwords list, quantity suffix words list and surname list are used for further precision improvement. The comparative evaluation with the famous Chinese word segmentation system ICTCLAS, using news and papers as testing text, shows that BUAASEISEG outperforms ICTCLAS in new words recognition and disambiguation.