基于内容的计算机科学论文推荐系统设计与实现

发布时间：2018-05-23 13:37

本文选题：推荐系统 + CHI特征选择　；参考：《吉林大学》2017年硕士论文

【摘要】：由于网络信息的巨大增长和复杂度的日益增加,用户很难从海量数据中准确地查找到他们想要的信息,尤其是对于研究者们来说,想要及时了解所研究领域的动态,有时并非易事。如今,世界上许多新知识主要以数字形式表现并存储在数字图书馆(Digital Libraries)系统中,所以数字图书馆正在进入一个黄金时代。这样的数据图书馆,在科技领域中,有如ACM Library、IEEE Library等。然而随着科技的发展和信息的不断增长,这些趋势的产生导致了一个不可避免的问题——信息过载。例如,研究者们想要选择合适的出版物发表论文时,发现大量的出版物与他们的搜索查询匹配,但是在很大程度上与它们的实际需求无关,这种情况让他们在选择上无所适从。故研究者们急需一个论文推荐系统,帮助他们选择合适的出版物。推荐系统是帮助用户从大量的数据中找到自己想要的信息的一个解决方案。它是一种信息过滤工具,主要根据用户的历史行为等信息来构建用户喜好模型,用于预测用户对物品的“评分”或“偏好”,并通过该喜好模型向用户推荐其他相关信息。推荐系统近年来非常流行,在各行各业得以应用。其对象主要包括:音乐、产品、新闻、搜索查询、学术论文等。为了帮助研究者选择合适的出版物并加快他们的投递进度,本文针对科技类出版物,设计并实现了一个基于内容的科技论文推荐系统。该系统基于CHI特征选择和Softmax回归的内容过滤推荐模型,用户只需要输入文章的摘要,系统即可向用户推荐该文章可发表的合适的出版物备选。该系统由三个部分组成:(1)数据爬取系统。(2)基于CHI的特征选择和Softmax回归的内容过滤推荐算法。(3)基于Web服务的科技论文推荐系统。由于目前国际上没有论文推荐系统相关数据集,且考虑到计算机科学领域的不断变化和扩展,以及论文集的每日更新,本文设计并研发了论文摘要自动爬取系统,该系统能够自动地持续更新训练数据。基于爬取到的数据,本文设计了基于CHI的特征空间生成方法,利用TF-IDF来计算每篇摘要的特征向量,然后使用Softmax回归对特征向量分类进而向用户推荐出版物。为了验证本文提出的基于内容的科技论文推荐算法的有效性,在本文数据爬虫爬取到的数据的基础上,分别对比了基于互信息(MI)的特征选择方法、基于信息增益(IG)的特征选择方法和基于卡方统计(CHI)的特征选择方法,并利用Softmax回归对特征向量分类,根据分类得分来判定推荐结果。从实验的比较结果得出,对于论文推荐系统来说,CHI模型和IG模型更适用于特征选择。最后,基于本文提出的推荐算法,本文设计并实现了基于Web服务的科技论文推荐系统,该推荐系统根据用户输入的摘要,分别向用户推荐三个期刊和三个会议,供用户投稿参考。
[Abstract]:Because of the huge increase of the network information and the increasing complexity of the network, it is difficult for users to find out exactly the information they want from the massive data, especially for the researchers, who want to keep abreast of the developments in the research field. Sometimes it is not easy. Nowadays, a lot of new knowledge in the world is mainly expressed in digital form and stored in digital library system, so digital library is entering a golden age. Such a data library, in the field of technology, such as the ACM Library IEEE Library, and so on. However, with the development of science and technology and the increasing of information, these trends lead to an inevitable problem-information overload. For example, when researchers want to choose the right publication to publish a paper, they find that a large number of publications match their search queries, but to a large extent have nothing to do with their actual needs, which leaves them at a loss in their choice. So researchers urgently need a paper recommendation system to help them select the right publications. Recommendation system is a solution to help users find the information they want from a large amount of data. It is a kind of information filtering tool, which mainly constructs the user preference model according to the information of the user's history and behavior, which is used to predict the user's "score" or "preference" on the items, and recommends other relevant information to the user through the preference model. Recommendation system is very popular in recent years, can be applied in various industries. Its main objects include: music, products, news, search queries, academic papers and so on. In order to help researchers select suitable publications and speed up their delivery, this paper designs and implements a content-based recommendation system for scientific and technological publications. The system is based on the CHI feature selection and Softmax regression content filtering recommendation model. Users only need to input the summary of the article, and the system can recommend the suitable publication options for the article. The system consists of three parts: 1) data crawling system. 2) feature selection based on CHI and content filtering recommendation algorithm based on Softmax regression. Because there are no related data sets of thesis recommendation system in the world at present, considering the constant change and expansion of computer science field and the daily update of the thesis collection, this paper designs and develops a paper abstract automatic crawling system. The system can automatically continuously update training data. Based on the crawling data, this paper designs a feature space generation method based on CHI, calculates the feature vector of each abstract by using TF-IDF, then classifies the feature vector by Softmax regression and recommends the publication to the user. In order to verify the validity of the proposed content-based recommendation algorithm for scientific and technological papers, based on the data crawled by the crawler in this paper, the feature selection methods based on mutual information (MIM) are compared, respectively. The feature selection method based on information gain (IGG) and the feature selection method based on chi-square statistics (Chi). The feature vector is classified by Softmax regression and the recommended result is judged according to the classification score. The experimental results show that the Chi model and the IG model are more suitable for feature selection in the thesis recommendation system. Finally, based on the recommendation algorithm proposed in this paper, this paper designs and implements a science and technology paper recommendation system based on Web service. The recommendation system recommends three journals and three conferences to users according to the abstracts of user input. For user contribution reference.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】