足球视频搜索引擎中的用户搜索信息获取与偏好挖掘

发布时间：2018-07-02 22:31

本文选题：足球视频 + 视频检索　；参考：《华中科技大学》2012年硕士论文

【摘要】：互联网信息量的急速增长使得人们淹没在信息的海洋中。尽管搜索引擎为用户提供了便捷的信息检索服务，但搜索引擎召回的成千上万的结果仍需要人们花费很大的精力与时间去浏览符合自身兴趣的信息。因此，从用户与搜索引擎的交互信息中挖掘用户偏好并为用户提供个性化的检索服务具有十分重要的意义。在系统分析了偏好挖掘的国内外研究现状的基础上，综合用户的显式反馈信息和隐式反馈信息实现了基于足球领域信息的用户偏好信息挖掘。针对偏好挖掘的实时性要求，通过对用户的检索信息进行语义分析确定用户检索会话的边界，以会话为单位获取隐式反馈信息为偏好挖掘提供实时的用户行为数据。对用户反馈信息进行分析，提取其中的偏好标签和偏好动作并将其描述为标签权重有向图，为偏好模型构的建提供数据。基于足球领域知识设计了分层权重无向图用户偏好模型，为用户偏好建模奠定基础。考虑到不同的偏好动作所代表的喜好程度不一样，对不同的偏好动作赋予不同的权重。结合历史偏好信息进行实时偏好挖掘并引入了时间衰减因子，将当前未出现的偏好信息的权值进行衰减，描述用户偏好的变化过程。将偏好挖掘算法应用于搜球网，为搜球网用户提供个性化的视频检索与视频推荐服务。实验结果表明，基于分层权重无向图模型的偏好挖掘算法能很好地从用户反馈信息中发掘用户的长期、中期以及短期的偏好信息。相比于原始的检索系统，基于用户偏好的个性化检索结果排序和视频推荐中起到了很好的效果，提高了搜球网的用户体验。但系统目前仅仅考虑文本查询信息，，尚未考虑用户提交的检索图片这一偏好信息来源。同时在偏好分析时未考虑偏好标签的修饰词对偏好挖掘的作用与影响。这两方面的内容将是未来研究的重点。
[Abstract]:The rapid growth of Internet information makes people submerged in the ocean of information. Although search engines provide users with convenient information retrieval services, the tens of thousands of results recalled by search engines still require people to spend a lot of energy and time to browse information that is in line with their own interests. Therefore, it is of great significance to mine user preferences from the interactive information between users and search engines and to provide personalized retrieval services for users. Based on the systematic analysis of the current research situation of preference mining at home and abroad, the user preference information mining based on football domain information is realized by integrating the explicit feedback information and implicit feedback information of users. According to the real-time requirement of preference mining, the boundary of user retrieval session is determined by semantic analysis of user's retrieval information, and real-time user behavior data is provided for preference mining with implicit feedback information obtained by session. The user feedback information is analyzed and the preference tags and preference actions are extracted and described as label weight digraphs to provide data for the construction of preference models. Based on football domain knowledge, a hierarchical weight undirected graph user preference model is designed, which lays a foundation for user preference modeling. Considering that different preference actions represent different degrees of preference, different preference actions are given different weights. This paper combines historical preference information with real-time preference mining and introduces time decay factor to attenuate the weights of current preference information and describe the process of user preference change. The preference mining algorithm is applied to the search net to provide personalized video retrieval and video recommendation services for the users. The experimental results show that the preference mining algorithm based on delamination weight undirected graph model can well extract the long-term, medium-term and short-term preference information from user feedback information. Compared with the original retrieval system, personalized retrieval result ranking and video recommendation based on user preference play a good role in improving the user experience. However, at present, the system only considers the text query information, and does not consider the user submitted image retrieval as a preferred information source. At the same time, the effect of preference tag modifier on preference mining is not considered in preference analysis. These two aspects of the content will be the focus of future research.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3

【参考文献】