结合概率潜在语义分析的文本谱聚类方法研究

摘要

传统谱聚类的相似矩阵建立在VSM(Vector Space Model)之上,该模型把词看作孤立的单元,没有考虑自然语言中存在大量的同义词、多义词现象.针对这一问题,提出一种用概率潜在语义分析(Probabilistic Latent Semantic Analysis,PLSA)来提取文本中隐含语义信息的方法,并构建文本集的相似矩阵,从语义的角度考虑了文本之间的相关性.实验结果表明,利用该方法得到的聚类精度有较大提高,结果要好于传统的谱聚类算法,从而验证了该方法的有效性.%Traditional similar matrix of spectral clustering is dependent on vector space model, which regards index word as independent unit and ignores a large number of synonyms and polysemy existing in natural language.To solve this prob lem, the paper comes up with a new method of extracting semantic information implicit in the text and constructing the simi lar matrix based on Probabilistic Latent Semantic Analysis(PLSA) .which takes into account the similarities of the texts.Ex periments indicate that such similar matrix built by PLSA can greatly improve categorization precision, and bring better re sults than traditional way like spectral clustering,further proves the availability of PLSA.