基于领域知识图谱的个性化推荐方法研究

发布时间：2019-04-30 12:41

【摘要】：在当下这个互联网技术飞速发展的时代,文档万维网己经转变成为语义网,语义网应用在各行各业的各个领域,而知识图谱是语义网的最为直观有效的表示,这也使得知识图谱的构建成为了当下的研究热点。尤其对于特定领域而言,实现个性化服务更加需要知识图谱来作为其坚定的基础,因此,近年来知识图谱在推荐系统的应用也成为了研究的重点。本文针对特定领域的知识图谱在构建过程中实体消歧环节和在领域知识图谱在信息推荐方面的应用,主要做了以下几个方面的研究工作。1.结合词向量和图模型的方法来实现实体消歧。针对特定领域知识图谱的构建提出了一种结合词向量和图模型的实体消歧方法。通过词向量计算工具Word2Vec构建词向量模型,结合人工标注的实体关系图谱,采用一种基于图的随机游走算法辅助计算相似度,使其能够较准确地计算旅游领域词与词之间的相似度。最后,提取待消歧实体的背景文本的若干关键词和知识库中候选实体文本的若干关键词,利用训练好的词向量模型结合图模型分别进行交叉相似度计算,把相似度均值最高的候选实体作为最终的目标实体。实验结果表明,这种新的相似度计算方法能够有效获取实体指称项与目标实体之间的相似度,从而能够较为准确地实现特定领域的实体消歧。2.基于属性图聚类的旅游领域个性化信息推荐。在前面的基础上构建领域实体的属性图,然后利用属性图聚类的方法进行用户的偏好发现,接着将领域实体划分成不同的旅游实体类别,再将用户的偏好信息结合领域实体的实体类别进行领域实体的属性图聚类,从而针对不同的用户做出相应的推荐,最后对这种基于属性图聚类的聚类推荐模型进行实验分析。3.旅游领域个性化信息推荐原型系统实现。本文将领域信息推荐算法用程序实现,通过抓取用户搜索的关键词来作为系统的输入,然后结合属性图聚类模型进行计算,最终将符合用户兴趣爱好的旅游景点展现给用户,实现领域知识的个性化推荐。
[Abstract]:In the current era of rapid development of Internet technology, the document World wide Web has been transformed into a semantic Web, and the semantic Web is applied in various fields in various industries, and the knowledge graph is the most intuitive and effective representation of the semantic Web. This also makes the construction of knowledge graph become a hot research topic at present. Especially for specific fields, the realization of personalized service needs knowledge graph as its firm foundation. Therefore, in recent years, the application of knowledge graph in recommendation system has also become the focus of research. Aiming at the application of entity disambiguation in the construction of domain-specific knowledge graph and the application of domain knowledge graph in information recommendation, this paper has done the following research work. 1. The method of combining word vector and graph model to realize entity disambiguation. In this paper, an entity disambiguation method combining word vector and graph model is proposed for the construction of domain-specific knowledge graph. The word vector model is constructed by word vector computing tool Word2Vec, and the similarity degree is calculated by a random walk algorithm based on graph, which is combined with the manually labeled entity relation graph. So that it can calculate the similarity between words in tourism domain more accurately. Finally, several keywords of the background text of the entity to be disambiguated and some keywords of the candidate entity text in the knowledge base are extracted, and the cross-similarity degree is calculated by using the trained word vector model combined with the graph model. The candidate entity with the highest similarity mean is regarded as the final target entity. The experimental results show that the new similarity calculation method can effectively obtain the similarity between the entity reference term and the target entity, so that the entity disambiguation in a specific field can be realized more accurately. Personalized information recommendation in tourism field based on attribute graph clustering. On the basis of the above, we construct the attribute graph of domain entity, then use the method of attribute graph clustering to discover the user's preference, and then divide the domain entity into different categories of tourism entities. Then the user's preference information is combined with the entity category of the domain entity to cluster the domain entity's attribute map, so as to make the corresponding recommendation for different users. Finally, this clustering recommendation model based on attribute graph clustering is experimentally analyzed. 3. Implementation of personalized information recommendation prototype system in tourism field. In this paper, the domain information recommendation algorithm is implemented by program, by grabbing the keywords of user search as the input of the system, and then combining the attribute graph clustering model to calculate, finally, the tourist attractions that accord with the user's interests will be displayed to the user. Implement personalized recommendation of domain knowledge.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】