Abstract: This paper proposes a Latent Document Similarity Model(LDSM). It denotes each document pair as a bipartite graph, where each node is a latent topic, and each edge is weighted with the similarity between the corresponding topics, and it represents the document similarity as the optimal matching of the bipartite graph. Experimental results show that LDSM outperforms the document similarity model based on TextTiling and the optimal matching of bipartite graph at both average precision and average recall.

Key words: topic, document similarity, document retrieval, information retrieval

摘要： 提出一种潜在文档相似模型(LDSM)，把每对文档看作一个二分图，把文档的潜在主题看作二分图的顶点，用主题间的加权相似度为相应边赋权值，并用二分图的最佳匹配表示文档的相似度。实验结果表明，LDSM的平均查准率和平均查全率都优于用TextTiling和二分图最佳匹配方法构建的文档相似模型。

关键词: 主题, 文档相似度, 文档检索, 信息检索

CLC Number: 

• TP311