郭理 张恒旭 王嘉岐 秦怀斌



摘 要: 由于大量新词的出现,使得中文文本分析产生了较大的困难,因此新词发现成为目前中文自然语言处理中的热点和难点问题。为此,文中提出了一种基于Trie树的词语左右熵和互信息新词发现算法。先根据成词规则,筛选掉文本中的停用词和非中文字符,将每个字与其右邻的字组成二元组;然后利用左右信息熵和互信息进行成词概率的计算,根据计算到的成词概率和词频筛选出新词;并且设计了三个实验,验证了算法的有效性和可行性。实验结果表明,该新词发现算法成词准确率较高,比其他新词发现算法时间效率有较大的提高,对于中文分词结果的优化起到重要的作用。
关键词: 新词发现算法; 左右熵; 互信息; Trie树; 算法设计; 对比验证
中图分类号: TN911?34; TP391.1 文献标识码: A 文章编号: 1004?373X(2020)06?0065?05
Trie tree based new word discovery algorithm using left?right entropy and
mutual information
GUO Li, ZHANG Hengxu, WANG Jiaqi, QIN Huaibin
(College of Information Science and Technology, Shihezi University, Shihezi 832000, China)
Abstract: The emergence of multitude of new words makes Chinese discourse analysis difficult. Therefore, the discovery of new words has become a hot and difficult problem in natural language processing of Chinese. A Trie tree based new word discovery algorithm using left?right entropy and mutual information is proposed. The disused words and non Chinese characters in the text are filtered out according to the rules of word?formation. Each word is divided into a binary group with its right neighbor, and then the probability of word?formation is calculated by means of left?right information entropy and mutual information, so as to screen out new words according to the calculated probability of word?formation and word frequency. Three experiments were designed to verify the effectiveness and feasibility of the algorithm. The experimental results show that the new word discovery algorithm has higher accuracy of word?formation, and has higher time efficiency than other new word discovery algorithms, which plays an important role in the optimization of Chinese word segmentation results.
Keywords: new word discovery algorithm; left?right entropy; mutual information; Trie tree; algorithm design; comparison validation
0 引 言
汉语言历史悠久,文化源远流长,很多地方有不同的方言,有着不同的书写习惯,这些方言大多不被包含在已有的词库中;另外,随着互联网的发展,产生了不同于传统语言的网络语言,其中包含着大量不存在于词库中的词,但是这些词却广泛存在于互联网中,并且常常是热门话题的关键词。由于中文无法像西方语言那样使用词与词间的分隔符进行分词,因此在面向中文的自然语言处理中,如何对中文文本进行分词处理就成为了一个难点。一个比较简单直接的解决方案是建立一个词库来进行中文词汇的分词处理。……