Chinese Organization Name Recognition Using Chunk Analysis

Chinese Organization Name Recognition Using Chunk Analysis Jihao YinP1 P, Xiaozhong FanP1 P, Kaixuan ZhangP2 P, Jiangde YuP1,3 P P1 P Department of Computer Science and Engineering, Beijing Institute of Technology, Beijing, 100081, China; P2 PLiaoning Technical University, Fuxin, Liaoning, 123000, China; P3 P Department of Computer Science, Anyang Teachers’ College, Anyang, Henan, 455000, China yin_jhgg@bit.edu.cn Abstract. A simplified N-best cascade model is put forward about Chinese organization names automatic recogni- tion. This model can do words segmentation, part-of-speech tag, chunks analysis and Chinese organization names recognition. The N-best cascade method can limit not only the errors propagation but also the search space. In the experiments, we integrate heuristic information and Organization name abbreviations processing into the model to achieve the better experiment results. The last precision and recall of Chinese organization names recognition are 92.31% and 81.01% in IEER99 newswire test set. Keywords: Chunk analysis; Chinese organization names recognition; N-best cascade model; Heuristic information 1 Introduction Named entity recognition is a foundational job in natural language processing, is an important pretreat- ment module too. Generally, the assignment of named entity recognition is identifying person names, location names, organization names, digital and time expressionP[1] P from a text. Person names, location names and organization names are not only difficult but also important. Many researches about person names and location names have been done, and their precision and re- call have meet practical need. But organization names recognition cannot satisfy need whether in theory or in practiceP[2] P. Many scholars have done some researches about Chinese chunk analysis. GuoDong Zhou et al identi- fied English named entities using an HMM-based chunk taggerP[3] P. But the study of Chinese organization names recognition using chunk analysis technology is a little. Noun-chunk is almost like the structure of Chinese organization names, so it is feasible to identify Chinese organization names using Noun-chunk analysis technology. Based on Chinese chunk analysis technology and name entity recognition, we put forward a simpli- fied N-best cascade model by mining the characters of Chinese organization names. In the experiments, we added heuristic information and organization name abbreviations processing, which improved the precision and recall of Chinese organization names recognition obviously. 2 Noun-chunk Automatic recognition 2.1 Noun-chunk Definition Noun-chunk usually includes noun and its former modifier. The former modifier includes number, ad- jective and noun, but excludes “的” structure. The modifier and center word of “的” structure fall into the different chunks separately[4]. (1) Center word of noun-chunk 347 The center words of noun-chunk must be nouns. According to the part-of-speech tag criterion of Beijing University, the center words include: noun (n), person name (nr), location name (ns), proper noun (nz), gerund (vn), noun element (Ng), noun abbreviation (j) etc. For example: [NP 中国/ns 宋庆龄/nr 基金会/n ]; [NP 中共/j 北京/ns 市委/n ]. (2) Modifier One noun-chunk cannot include the other noun-chunk. The modifier must include: number phrase, ad- jective phrase, noun, pronoun. For example: [NP 第一/m 汽车/n 制造/vn 厂/Ng] 2.2 HMM-based Chunk Analysis Hidden Markov Model (HMM) is simple and …

Content truncated to 500 words