排序主题模型及其应用研究

项目名称： 排序主题模型及其应用研究

项目编号： No.61272369

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 鲁明羽

作者单位： 大连海事大学

项目金额： 80万元

中文摘要： 主题模型是近年来兴起的可以从离散数据集中自动提取隐含语义结构的一种生成概率模型，是从海量数据中发现隐含语义主题的有效工具，可广泛应用于信息检索、自动文摘及推荐系统等领域。本项目遵循"小题精做"的原则，对一类重要的主题模型- - 排序主题模型进行深入研究,主要针对两种排序主题模型开展算法及其应用基础研究：第一种是查询无关的排序主题模型，通过度量词项之间的关系并结合主动学习方法，解决没有查询词情况下获得有序主题分布的问题；第二种是面向查询的排序主题模型，借鉴排序学习的思想，解决存在查询词情况下得到有序主题分布的问题。其次，本项目还将进行主题模型的参数推断方法研究，力求在提高推断精度的同时尽可能降低计算复杂度。最后，本项目拟将所提出的排序主题模型应用于多文本自动文摘和论文推荐系统中，研制开发两套应用系统，达到既验证算法又促进算法实用化的目的。

中文关键词： 机器学习；排序主题模型；多文档自动文摘；推荐系统；脑认知

英文摘要： Topic model is a kind of probabilistic model which can extract latent semantic structure from discrete data, and is effective to discover hidden meaningful topic from very large scale data. It can be applied widely in information retrieval, automatic summarization and recommender systems. Bearing the principle of focussing small topics, this project intents to research on one particular important topic model - - - ranking topic model. Two kinds of ranking topic models are proposed and to be applied in real-world applications: query-independent ranking topic models intent to get the ranked topic distributions without query words by evaluating relationships between terms and using active learning methods, and query-dependent ranking topic models intent to solve the problem of getting ranked topic distribution with query words by adopting learning-to-rank methods. Meanwhile, the project also intent to propose a new parametric inference method to enhance the model performance, which aims to reduce the computation complexity and raise the inference accuracy. At last, the project will apply the proposed ranking topic models to multi-document automatic summarization and academic paper recommender system. With two system established, the algorithms proposed in the project are evaluated and applied.

英文关键词： Machine Learning；Ranking Topic Model；Multi-document Automatic Summarization；Recommender System；Brain cognition