基于混合模型的学术论文推荐方法研究
发布时间：2018-10-31 21:00
【摘要】：近年来随着互联网技术的高速发展,学术研究领域也发生着翻天覆地的变化,网络上学术论文的数量呈爆炸式增长。研究人员在网络上查找其所需要的学术论文信息时,往往需要花费大量的时间和精力,因此如何快速、准确的为研究人员找到其感兴趣的学术论文信息成为亟待解决的问题。 本文主要围绕研究人员学术研究兴趣建模,以及如何准确地向研究人员推荐学术论文展开研究。论文在对基于内容推荐算法中的主题模型和协同过滤方法中的模型推荐方法研究的基础上,融合两种推荐方法提出了一种新的混合推荐方法,改善了协同过滤推荐方法中数据稀疏性对于推荐效果的不良影响。本文编码实现了提出的基于混合模型的学术论文推荐方案,通过实验确定了方案中的一些参数取值,并与其他推荐方案进行了对比分析,验证了本方案的有效性和优势。 本文提出的方案包括一种新的主题模型—-ACTOT(Author Conference Topic Over Time)以及基于该模型的混合推荐模型MFWT (Matrix Factorization With Topic)。ACTOT模型结合了论文的内容信息、发表期刊／会议信息和发表时间信息,可以准确地对研究人员的兴趣进行建模。MFWT (Matrix Factorization With Topic)模型在实现了基于模型的协同过滤方法和基于内容的推荐方法的混合,使用ACTOT模型和LDA模型计算的用户主题向量和论文的主题向量,并分别对PMF(Probabilistic Matrix Factorization)模型中的用户隐式因子特征向量和论文隐式因子特征向量作正则化处理,修正了PMF模型的推荐结果,有效地改善了评分矩阵稀疏性带来的不良影响,同时也解决了协同过滤方法的冷启动问题。 本文首先分析了学术研究领域现在主流推荐方法的研究现状和不足之处,然后详细介绍了本文提出的MFWT混合模型设计方案和实现方法,最后介绍了MFWT模型的实验验证和实验结果分析。
[Abstract]:In recent years, with the rapid development of Internet technology, the academic research field has also undergone earth-shaking changes, the number of academic papers on the network explosive growth. Researchers often spend a lot of time and energy when searching the information of academic papers they need on the network. Therefore, how to find the information of academic papers of interest to researchers quickly and accurately becomes an urgent problem to be solved. This paper focuses on the modeling of researchers' interest in academic research and how to recommend academic papers to researchers accurately. Based on the research of topic model in content-based recommendation algorithm and model recommendation method in collaborative filtering method, a new hybrid recommendation method is proposed. It improves the bad effect of data sparsity on recommendation effect in collaborative filtering recommendation method. In this paper, the proposed scheme of academic thesis recommendation based on hybrid model is implemented, and some parameters of the scheme are determined by experiments, and compared with other schemes, the effectiveness and advantages of this scheme are verified. The scheme proposed in this paper includes a new topic model, ACTOT (Author Conference Topic Over Time), and a hybrid recommendation model based on this model, MFWT (Matrix Factorization With Topic). ACTOT model, which combines the content information of the paper. Publishing journal / conference information and publishing time information can accurately model the. MFWT (Matrix Factorization With Topic) model of researchers' interest in implementing a mixture of model-based collaborative filtering methods and content-based recommendations. Using the user topic vector calculated by ACTOT model and LDA model and the topic vector of the paper, the user implicit factor eigenvector and the paper implicit factor feature vector in PMF (Probabilistic Matrix Factorization) model are regularized, respectively. The recommended results of PMF model are corrected to improve the bad effect of score matrix sparsity and the cold start problem of collaborative filtering method is also solved. This paper first analyzes the research status and shortcomings of the current mainstream recommendation methods in the academic research field, and then introduces the design scheme and implementation method of the MFWT hybrid model proposed in this paper in detail. Finally, the experimental verification of MFWT model and the analysis of experimental results are introduced.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3
本文编号：2303534
[Abstract]:In recent years, with the rapid development of Internet technology, the academic research field has also undergone earth-shaking changes, the number of academic papers on the network explosive growth. Researchers often spend a lot of time and energy when searching the information of academic papers they need on the network. Therefore, how to find the information of academic papers of interest to researchers quickly and accurately becomes an urgent problem to be solved. This paper focuses on the modeling of researchers' interest in academic research and how to recommend academic papers to researchers accurately. Based on the research of topic model in content-based recommendation algorithm and model recommendation method in collaborative filtering method, a new hybrid recommendation method is proposed. It improves the bad effect of data sparsity on recommendation effect in collaborative filtering recommendation method. In this paper, the proposed scheme of academic thesis recommendation based on hybrid model is implemented, and some parameters of the scheme are determined by experiments, and compared with other schemes, the effectiveness and advantages of this scheme are verified. The scheme proposed in this paper includes a new topic model, ACTOT (Author Conference Topic Over Time), and a hybrid recommendation model based on this model, MFWT (Matrix Factorization With Topic). ACTOT model, which combines the content information of the paper. Publishing journal / conference information and publishing time information can accurately model the. MFWT (Matrix Factorization With Topic) model of researchers' interest in implementing a mixture of model-based collaborative filtering methods and content-based recommendations. Using the user topic vector calculated by ACTOT model and LDA model and the topic vector of the paper, the user implicit factor eigenvector and the paper implicit factor feature vector in PMF (Probabilistic Matrix Factorization) model are regularized, respectively. The recommended results of PMF model are corrected to improve the bad effect of score matrix sparsity and the cold start problem of collaborative filtering method is also solved. This paper first analyzes the research status and shortcomings of the current mainstream recommendation methods in the academic research field, and then introduces the design scheme and implementation method of the MFWT hybrid model proposed in this paper in detail. Finally, the experimental verification of MFWT model and the analysis of experimental results are introduced.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3
【参考文献】
相关期刊论文 前2条
1 王国霞;刘贺平;;个性化推荐系统综述[J];计算机工程与应用;2012年07期
2 陈树年;搜索引擎及网络信息资源的分类组织[J];图书情报工作;2000年04期
，本文编号：2303534
本文链接：https://www.wllwen.com/wenshubaike/lwzy/2303534.html
