基于合著网络的论文混合推荐算法研究

发布时间：2019-04-04 18:03

【摘要】：科技文献的高速增长使得科研信息的检索难度大大增加，虽然搜索引擎在很大程度上减轻了科研人员检索论文的工作，但它缺少对科研人员个性化需求的考量，难以在搜索结果中进一步找到与其兴趣相关内容，而推荐系统能够有效解决这些问题。本文首先介绍了常见推荐算法及其优缺点，然后介绍了合著网络分析方法和技术，研究了社会学中的科研合作现象，分析了网络整体特征和节点重要性，验证了合著网络的复杂网络特性，解释了学者的社会性和团体性。单一推荐算法由于自身缺陷和应用限制，在论文推荐效果上并不理想。根据研究现状及不足，本文从以下几方面研究混合推荐算法的设计： 1.为准确描述用户兴趣，，用已发表论文相关信息构建动态用户兴趣模型，同时用论文质量评价方法描述论文重要性，在二者的基础上提出一种混合推荐算法； 2.为减少推荐的盲目性，将社会学中的合著网络引入混合推荐算法中，定义了不同用户之间合作强度计算方式，对合著网络进行社团划分以限制合作强度传播范围； 3.用户对排名靠前的论文具有高阅读倾向，为衡量混合推荐算法对结果的排序能力，引入信息检索系统评价指标—平均准确率和平均排序倒数对Top-N推荐效果进行评价。实验结果表明混合推荐算法相对于单一推荐算法具有较优的推荐效果，引入社团划分的混合推荐算法具有更优的推荐效果。
[Abstract]:The rapid growth of scientific and technological literature has greatly increased the difficulty of searching scientific research information. Although the search engine has greatly alleviated the work of searching papers for scientific researchers, it lacks the consideration of the individual needs of scientific researchers. It is difficult to find the content related to its interest in the search results, and the recommendation system can effectively solve these problems. This paper first introduces the common recommendation algorithms and their advantages and disadvantages, then introduces the co-author network analysis methods and techniques, studies the phenomenon of scientific research cooperation in sociology, analyzes the overall characteristics of the network and the importance of nodes. The complex network characteristics of co-authored networks are verified, and the sociality and collectivity of scholars are explained. Single recommendation algorithm is not ideal because of its own defects and application limitations. According to the present situation and deficiency of the research, this paper studies the design of hybrid recommendation algorithm from the following aspects: 1. In order to accurately describe user interest, a dynamic user interest model is constructed with relevant information of published papers. At the same time, the paper quality evaluation method is used to describe the importance of the paper, on the basis of which a hybrid recommendation algorithm is proposed. In order to reduce the blindness of recommendation, the co-author network in sociology is introduced into the hybrid recommendation algorithm, and the calculation method of cooperation intensity between different users is defined, and the co-author network is divided into communities to limit the spread range of cooperation intensity; 3. Users have a high reading tendency to the top papers. In order to measure the ability of hybrid recommendation algorithm to sort the results, the evaluation index of information retrieval system-average accuracy and the reciprocal of average ranking are introduced to evaluate the effect of Top-N recommendation. The experimental results show that the hybrid recommendation algorithm has a better recommendation effect than a single recommendation algorithm, and the hybrid recommendation algorithm based on community partition has a better recommendation effect.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【参考文献】