【作者】 王博；

【导师】 何明；

【作者基本信息】 北京工业大学 ， 计算机技术（专业学位）， 2020， 硕士

【摘要】 随着信息技术的高速发展和互联网规模的日益扩大,产生了海量的数据信息,人们从中获取所需要的信息面临着巨大的挑战。为解决上述难题,个性化推荐技术应运而生,它能够主动向用户推荐其具有潜在兴趣的项目。协同过滤是目前主流的个性化推荐技术之一,已经广泛应用于电子商务、社交媒体等互联网诸多领域,并取得了显著效果。然而,在面临数据稀疏性和冷启动问题时,其推荐性能往往会急剧下降。近年来,知识图谱技术引起了学术界和工业界的广泛关注,已被应用于推荐系统和语义搜索等领域并取得重要突破。由于知识图谱中含有丰富的语义信息,因此通过将其与协同过滤推荐算法相结合,能够有效解决传统协同过滤推荐算法中存在的数据稀疏性和冷启动问题,从而提升推荐系统的性能。然而,现有的基于知识图谱的推荐技术更多地关注知识库中的项目属性信息,并未充分考虑用户属性信息。同时,尚有大量与知识有关的异构信息没有得到有效利用,限制了推荐性能的进一步提升。针对上述问题,本文对面向领域的知识图谱构建技术、多源信息融合的知识表示学习方法、以及知识图谱在个性化推荐中的应用进行了研究,主要工作如下:(1)构建了面向电影领域的知识图谱。首先将外部大规模公共知识图谱中的实体与推荐系统数据集中的项目进行链接,然后将链接到的电影实体与推荐系统中项目信息、用户信息以及用户和项目间的隐式反馈等异构信息进行融合。最后将这些信息统一用知识图谱三元组形式进行结构化表示。(2)建立了电影知识图谱中项目及用户信息的嵌入表示。通过基于翻译模型的知识表示学习方法,将电影知识图谱中的用户、项目以及它们之间的关系嵌入到低维稠密的向量空间进行有效表示。(3)提出了基于知识图谱的协同过滤推荐算法。将知识图谱中丰富的语义知识作为辅助信息融入到协同过滤算法,解决了传统协同过滤推荐算法中的数据稀疏性和冷启动问题。(4)对本文提出的方法与其他几种典型的推荐算法进行实验评估与对比。实验结果表明本文提出的推荐算法在准确率、召回率、F1值和NDCG性能指标上优于其它算法。更多还原

【Abstract】 With the rapid development of information technology and the growing scale of the Internet,a large amount of data information has been generated,and people are facing great challenges in obtaining the information they need.In order to solve the above problems,personalized recommendation technology came into being,which can actively recommend projects of potential interest to users.As an important means to address the problem of information overload,personalized recommendation can actively recommend items of potential interest to users.Collaborative filtering recommendation is one of the mainstream personalized recommendation technologies,which has been widely used in e-commerce,social media and many other Internet fields,and has achieved remarkable results.However,in the face of data sparsity and cold start problems,its recommendation performance often drops sharply.In recent years,knowledge graph technology has attracted extensive attention of academia and industry,and has been widely used in recommendation system,semantic search and other fields and made important breakthroughs.Because knowledge graph contains rich semantic information,it can effectively solve the problem of data sparsity and cold start in traditional collaborative filtering recommendation algorithm by combining it with collaborative filtering recommendation algorithm,so as to improve the performance of recommendation system.However,the existing recommendation technology based on knowledge graph pays more attention to the project attribute information in the knowledge base,and does not fully consider the user attribute information.At the same time,a large number of heterogeneous information related to knowledge has not been effectively utilized,which limits the further improvement of recommendation performance.In view of the above problems,this thesis studies the domain oriented knowledge graph construction technology,the knowledge representation learning method of multi-source information fusion,and the application of knowledge graph in personalized recommendation.The main work is as follows:1.The knowledge graph of film field is constructed.Firstly,the entities in the external large-scale public knowledge graph are linked to the projects in the data set of the recommendation system,and then the linked movie entities are fused with the project information,user information and the implicit feedback information between users and projects in the recommendation system.Finally,this information is unified in the form of knowledge graph triplets for structured representation.2.The embedded representation of project and user information in film knowledge graph is established.Through the knowledge representation learning method based on translation model,users,projects and their relationships in the knowledge graph of movies are embedded into the low-dimensional dense vector space for effective representation.3.The recommendation algorithm combining knowledge graph and collaborative filtering is proposed.The rich semantic knowledge in knowledge graph is integrated into collaborative filtering algorithm as auxiliary information,which solves the problem of data sparsity and cold start in traditional collaborative filtering recommendation algorithm.4.The method proposed in this paper is compared and evaluated of experiments with other typical algorithms.Experimental results show that the proposed algorithm is superior to other algorithms in Precison,Recall, F1 and NDCG performance.更多还原