基于中心化相似度矩阵的词向量方法

词向量使用低维稠密向量表示词，通过向量运算能够反映词间关系，被广泛应用于自然语言处理任务。对基于矩阵分解的词向量方法进行了研究，发现降维前相似度矩阵质量与词向量质量存在线性相关性，提出了一种基于中心化相似度矩阵的方法。该方法使得相似（不相似或弱相似）词间的相似程度相对增强（减弱）。在WS-353和RW数据集的词语相似性实验中验证了所提出方法的有效性，两个数据集下词向量质量最高提升0.2896和0.1801。中心化能够提升降维前相似度矩阵质量，进而提升词向量质量。

Method of word vector based on centring similarity matrix

The word vector, which represents word by a low dimensional dense vector. The relationships between words are denoted by vector operations. Hence it is broadly applied in tasks of natural language processing. The method of word vector based on matrix factorization is studied. It found that there is a linear correlation between the quality of no dimension reduction matrix and the quality of word vector. Furthermore, it derived a method of the word vector, which based on a kind of centring similarity matrix. This method makes the similarity between similar (dissimilar or weakly similar) words relatively enhanced (weakened) . In the word similarity experiments of WS-353 and RW datasets, the effectiveness of the proposed method is verified. The highest quality of the word vectors among the two datasets is 0.2896 and 0.1801. Centralization can improve the quality of similarity matrix, moreover it can improve the quality of word vector.