个性化房地产信息检索与推荐技术研究

发布时间：2019-04-25 20:58

【摘要】：随着互联网上房地产信息越来越多，人们从大量的房产信息中找到自己需要的信息也变得越来越困难，因为传统的房地产信息检索系统中没有考虑用户的兴趣偏好，只是机械地进行信息搜索，对于不同的用户，，相同的查询词，往往返回相同的查询结果，这样的查询结果已经无法满足用户日益鲜明的个性化需求。本文针对目前房地产信息产业中的上述问题，提出了基于用户偏好模型的个性化房地产信息检索与推荐，并实现了个性化房地产信息检索与推荐系统。本文主要在以下几个方面开展了探索性研究工作：首先在对国内外相关文献收集整理的基础上，针对目前房地产信息中存在的问题，结合信息检索技术、用户模型技术和推荐技术，提出了课题研究的基本思路和技术路线。其次，对用户偏好模型展开深入研究。在模型的更新方面，通过对查询词和检索结果的操作行为进行学习，提出了适应度函数来指导模型的更新；在模型的表示方面，结合房产信息本体的特征采用向量空间模型表示，并设计了特征词典，实现了查询词到用户偏好的映射并能得到具体的权值，以便于相似度的计算来优化检索结果的排序。然后，在前面实现了用户偏好模型的基础上，分析了基于内容的推荐和协同过滤算法的优缺点，提出了一种自适应推荐算法，将基于内容的推荐和基于Item的协同过滤相结合，增加了一个数据结构——最近认可队列，为两种推荐算法的合并策略提供依据并能自适应地改变合并策略的参数，从而达到更好的推荐服务质量。最后，利用前面的理论知识，实现了一个个性化房地产信息检索与推荐系统。通过P@N评测指标对系统的检索性能进行测试，实验证明个性化房地产信息检索与传统的房地产信息检索相比性能有了显著改进，尤其是在对用户偏好模型做初始化的情况下改进效果更为明显，P@N评测指标提高了约25%。在前面实验的基础上，以Precision-Recall和覆盖率为指标对系统的推荐功能进行了测试，实验证明本文中提出的基于内容推荐和基于Item协同过滤相结合的自适应推荐算法相比原协同过滤算法性能有所提高。
[Abstract]:With more and more real estate information on the Internet, it is becoming more and more difficult for people to find the information they need from a large number of real estate information, because the traditional real estate information retrieval system does not consider the interests and preferences of users. Just carry on the information search mechanically, for different users, the same query words, often return the same query results, such query results have been unable to meet the user's increasingly distinct personalized needs. In view of the above-mentioned problems in the real estate information industry, this paper proposes a personalized real estate information retrieval and recommendation system based on user preference model, and implements a personalized real estate information retrieval and recommendation system. This paper mainly carried out exploratory research work in the following aspects: firstly, based on the collection and arrangement of relevant documents at home and abroad, aiming at the problems existing in the current real estate information, combined with the information retrieval technology, User model technology and recommendation technology, put forward the basic ideas and technical route of the subject research. Secondly, the user preference model is deeply studied. In the aspect of model updating, the fitness function is proposed to guide the updating of the model by learning the operation behavior of query words and retrieval results. In the representation of the model, the vector space model is used to represent the features of the real estate information ontology, and a feature dictionary is designed to realize the mapping of query words to user preferences and to obtain the specific weights. In order to optimize the ranking of retrieval results, the similarity calculation is used to optimize the order of retrieval results. Then, based on the previous implementation of user preference model, the advantages and disadvantages of content-based recommendation and collaborative filtering algorithm are analyzed, and an adaptive recommendation algorithm is proposed, which combines content-based recommendation with Item-based collaborative filtering. A data structure, the recently recognized queue, is added, which provides the basis for the merging strategy of the two recommendation algorithms and adaptively changes the parameters of the merge strategy, so as to achieve better recommended quality of service. Finally, using the previous theoretical knowledge, a personalized real estate information retrieval and recommendation system is implemented. The retrieval performance of the system is tested by PNN evaluation index. The experiment proves that the performance of personalized real estate information retrieval is significantly improved compared with the traditional real estate information retrieval. Especially in the case of initialization of the user preference model, the improvement effect is more obvious, and the PN evaluation index is increased by about 25%. On the basis of the previous experiments, the recommended function of the system is tested based on Precision-Recall and coverage. Experiments show that the performance of the proposed adaptive recommendation algorithm based on content recommendation and Item collaborative filtering is better than that of the original collaborative filtering algorithm.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【参考文献】