关键词: 排序学习, 构件检索, 潜在语义分析, 最大似然估计

Abstract: This paper applies the method of learning to rank to the research of component retrieval. Firstly, the facet description method is used to describe the components comprehensively, and the features of the facet described components are extracted through the Word2vec model and the weight setting method. Secondly, the component semantics analysis and cosine similarity calculation are performed on the component feature description information to obtain the component training data set. Finally, the component training data set and the component data set are used to train the model parameters of the improved Plackett-Luce probabilistic ranking model through the maximum likelihood estimation method, so as to obtain a component ranking model. The component ranking model is applied to the component retrieval to realize a component retrieval method. Experiments show that the method has better effectiveness, recall, precision and efficiency are better than the traditional component retrieval methods.

Key words: learning to rank, component retrieval, latent semantic analysis, maximum likelihood estimation