桌面搜索（或称个人信息检索）是定义在本地个人信息空间上的搜索过程，旨在帮助个人用户有效的搜索到所需要的本地资源（即文件）。近年来随着社会信息化进程的不断推进，大数据时代悄然来临，个人用户在本地计算机上生成和存储的数据爆炸式的增长。个人数据的存储和管理也已经进入了TB级时代。个人计算机用户对快速准确的搜索庞大的本地数据的需求日趋强烈。这使得桌面搜索在近年来成为了工业界和学术界关注和研究的热点领域。在工业界已经有一些被广大用户熟知的桌面搜索解决方案，比如Google Desktop Search、Windows Desktop Search等，这些传统的桌面搜索解决方案实现的是基于关键字的检索，而没有考虑本地资源之间潜在的语义关系。这就要求用户必须准确的记忆和键入搜索关键词，而这样的搜索结果其实是不充分的。在信息检索中，丰富的有意义的关联关系和信号的引入能够有效的提升搜索结果的质量。在本地环境下，资源之间直观上来看相互独立毫无关联。然而个人计算机上的资源的创建、浏览、存储因人而异，和用户的使用习惯、个人经验和记忆等息息相关。用户这种管理资源的习惯、经验和记忆在资源之间无形的产生了某些隐性的语义关联。对资源之间潜在的关联关系进行挖掘和利用为桌面搜索的研究提供了非常多的可能性。通过观察我们发现用户在使用个人计算机时有一个普遍的模式：“操作某些资源以完成跟某个特定主题相关的任务，并且这些资源会被用户根据资源之间的某种关系组织到某些特定的目录中存储”。这一发现启发我们“主题信息”、“用户历史行为信息”、“目录结构”对于定位本地资源是非常有帮助的。

Desktop Search refers to the process of searching within one’s personal space of information, which is aimed to help user search the local resources effectively. With the development of informatization, the personal data generated and stored on PC grows rapidly. The management on personal data steps into the era of TB. How to pinpoint the local resources among local data ocean quickly and actually has become a hot research topic in both industry and research communities. Major Internet service providers have released their prominent desktop search applications recently, such as Google Desktop, Windows Desktop. These traditional solutions are keyword based search without considering any kinds of implicit semantic relations, which will result in an insufficient search. In Information retrieval, introducing rich meaningful association signals can help improve the search. Intuitively, it seems that personal resources are independent with each other. In fact, most local items have been explicitly viewed, created, or saved by the user. As such, there items are personal to the individual and are intertwined with personal experience and memories which indicates that implicit associations among local resources exist extensively. These associations can be further used to improve traditional keyword-based search. We observe that users usually operate PC in a common pattern: Operating some resources to finish a specific task related to a certain topic, and organizing these resources in some directories. This observation inspires us that topic, user behaviors and directory structure are quite useful information for locating resources.

In this thesis, we propose a personal information retrieval approach based on a unified multi-dimensional latent semantic graph. The approach exploits the three kinds of information to improve traditional desktop search. We denote the three implicit information as {Task, Topic, Location} Relations respectively. The heart of our approach is Latent Semantic Graph (LSG), which is used to measure the three relations with associated score. Based on LSG, we develop a personalized rank schema to improve tradition keyword-based desktop search and design a creative semantic recommendation algorithm to expand the query results. We implement the prototype system based on LSG and conduct user experiments. Experiments reveal that the performance of proposed approach is superior to that of traditional keyword-based desktop search and our approach is effective.