社区问答用户兴趣建模方法研究

发布时间：2018-05-05 23:49

本文选题：社区问答系统 + 用户标签　；参考：《昆明理工大学》2017年硕士论文

【摘要】：随着Web 2.0技术的应用和发展,社区问答系统(CQA)作为新型知识获取模式中的一种典型应用,增强了提问者与回答者之间的交互性,使用户之间的联系更加紧密,同时用户可以对问答平台中已有的知识信息进行搜索和评价,这种应用相较于搜索引擎来说,无疑是一种新的信息获取模式,并且使用起来更加便捷有效。然而,伴随着社区问答系统的不断发展,其吸引了大量的用户参与,这使得在社区问答中的信息资源越来越庞大,并且由于社区问答的特有交互模式,使得CQA面临了许多问题和挑战,那么如何准确理解用户行为与需求,向用户提供完善的个性化推荐服务就成为了 CQA所亟待解决的问题。用户兴趣建模作为个性化推荐服务中的关键技术,用户兴趣模型的使用直接关系到系统个性化推荐服务的效果。本文针对社区问答中用户的兴趣模型建模方法进行了分析研究,主要完成了以下研究工作:(1)对以“百度知道”为例的社区问答结构进行了具体分析。在对社区问答用户的兴趣进行建模前,我们需要充分了解用户所处的使用环境。通过分析,我们可以了解社区问答用户在社区中的互动过程,对社区问答用户的需求有了一定的认知,为我们较好的获取用户兴趣信息提供了一定基础。(2)针对社区问答用户标签具有的特殊属性,提出了在社区问答用户兴趣建模的过程中融合用户标签的建模方法,该方法在用户兴趣特征提取过程中上融合了标签的作用,有效解决了用户兴趣特征集的稀疏性问题,增强了用户的兴趣特征表示,并且在建模过程中选用了谱聚类算法对模型进行学习,最终构建了融合标签的社区问答用户兴趣模型。实验结果表明,经过融合标签特征后可以有效提高用户兴趣模型的准确性,并且谱聚类算法在模型学习上的效果要优于传统的K-means聚类算法。(3)针对社区问答用户兴趣偏移的问题,对用户兴趣变化的特点进行了分析,基于艾宾浩斯遗忘规律的理论基础,采用了兴趣遗忘函数对用户兴趣权重做时间优化更新计算,构建了基于遗忘规律的社区问答用户兴趣模型,实现了用户兴趣模型的动态更新。实验结果表明,基于遗忘规律的社区问答用户兴趣建模方法在评价指标上要优于融合标签的社区问答用户兴趣建模方法。
[Abstract]:With the application and development of Web 2.0 technology, the community Q & A system is a typical application in the new knowledge acquisition mode, which enhances the interaction between the questioner and the responder, and makes the connection between users closer. At the same time, users can search and evaluate the existing knowledge information in the Q & A platform. Compared with the search engine, this application is undoubtedly a new mode of information acquisition, and it is more convenient and effective to use. However, with the continuous development of the community question and answer system, it attracts a large number of users, which makes the information resources in the community question and answer more and more huge, and because of the unique interactive mode of community question and answer. CQA is faced with many problems and challenges, so how to accurately understand the user's behavior and needs, and provide users with a perfect personalized recommendation service has become a problem to be solved urgently in CQA. User interest modeling is a key technology in personalized recommendation service. The use of user interest model is directly related to the effect of personalized recommendation service. In this paper, the modeling method of user's interest model in community Q & A is analyzed, and the following research work is done: 1) the community question and answer structure which takes "Baidu know" as an example is analyzed concretely. Before modeling the interest of community Q & A users, we need to fully understand the user's use environment. Through analysis, we can understand the interactive process of community Q & A users in the community, and have a certain understanding of the needs of community Q & A users. This paper provides a basis for us to obtain user interest information. (2) aiming at the special attributes of community Q & A user tags, a modeling method of integrating user tags in the process of interest modeling of community Q & A users is proposed. This method combines the functions of tags in the process of user interest feature extraction, solves the sparse problem of user interest feature set effectively, and enhances the user interest feature representation. In the process of modeling, the spectral clustering algorithm is used to learn the model, and finally, the interest model of community Q & A users is constructed. The experimental results show that the accuracy of user interest model can be improved by merging tag features, and the spectral clustering algorithm is better than the traditional K-means clustering algorithm in model learning. The characteristics of user interest change are analyzed. Based on the theoretical basis of Ibinhaus' forgetting law, the interest forgetting function is used to calculate the time optimal updating of user interest weight. The community Q & A interest model based on forgetting law is constructed, and the dynamic update of user interest model is realized. The experimental results show that the interest modeling method based on forgetting law is superior to the one based on fusion label in evaluating the interest of community Q & A users.
【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】