Joint Learning of Chinese Words, Terms and Keywords

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1774–1778, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Joint Learning of Chinese Words, Terms and Keywords Ziqiang Cao1 Sujian Li1 Heng Ji2 1Key Laboratory of Computational Linguistics, Peking University, MOE, China 2Computer Science Department, Rensselaer Polytechnic Institute, USA {ziqiangyeah, lisujian}@pku.edu.cn jih@rpi.edu Abstract Previous work often used a pipelined framework where Chinese word segmen- tation is followed by term extraction and keyword extraction. Such framework suf- fers from error propagation and is un- able to leverage information in later mod- ules for prior components. In this paper, we propose a four-level Dirichlet Process based model (DP-4) to jointly learn the word distributions from the corpus, do- main and document levels simultaneously. Based on the DP-4 model, a sentence-wise Gibbs sampler is adopted to obtain proper segmentation results. Meanwhile, terms and keywords are acquired in the sampling process. Experimental results have shown the effectiveness of our method. 1 Introduction For Chinese language which does not contain ex- plicitly marked word boundaries, word segmenta- tion (WS) is usually the first important step for many Natural Language Processing (NLP) tasks including term extraction (TE) and keyword ex- traction (KE). Generally, Chinese terms and key- words can be regarded as words which are repre- sentative of one domain or one document respec- tively. Previous work of TE and KE normally used the pipelined approaches which first conducted WS and then extracted important word sequences as terms or keywords. It is obvious that the pipelined approaches are prone to suffer from error propagation and fail to leverage information for word segmentation from later stages. Here, we provide one example in the disease domain, to demonstrate the common prob- lems in current pipelined approaches and propose the basic idea of our joint learning of words, terms and keywords. Example: @�•Ï�Ç(thrombocytopenia) �(with) {• (heparinoid) �(have) sû(relation). This is a correctly segmented Chinese sen- tence. The document containing the example sen- tence mainly talks about the property of ”{• (heparinoid)” which can be regarded as one key- word of the document. At the same time, the word @�•Ï�Ç(thrombocytopenia) appears fre- quently in the disease domain and can be treated as a domain-specific term. However, for such a simple sentence, current segmentation tools perform poorly. The segmen- tation result with the state-of-the-art Conditional Random Fields (CRFs) approach (Zhao et al., 2006) is as follows: @�•(blood platelet) Ï�(reduction) Ç(symptom) �{(of same kind) •(liver) �(always) sû(relation) where @�•Ï�Ç is segmented into three com- mon Chinese words and {• is mixed with its neighbors. In a text processing pipeline of WS, TE and KE, it is obvious that imprecise WS results will make the overall system performance unsatisfy- ing. At the same time, we can hardly make use of domain-level and document-level information col- lected in TE and KE to promote the performance of WS. Thus, one question comes to our minds: can words, terms and keywords be jointly learned with consideration of all the information …

Content truncated to 500 words