The importance of supporting keyword searches on XML data has been widely recognized. Different from structured queries, keyword searches are inherently ambiguous due to the inability/unwillingness of users to specify pinpoint semantics. As a result, processing keyword searches involves many unique challenges. In this paper we discuss the motivation, desiderata and challenges in supporting keyword searches on XML data. Then we present an XML keyword search engine, XSeek, which addresses the

more »

... lenges in several aspects: identifying explicit relevant nodes, identifying implicit relevant nodes, and generating result snippets. At last we discuss the remaining issues and future research directions. Introduction Information search is an indispensable component of our lives. Due to the vast collections of XML data on the web and in enterprises, providing users with easy access to XML data is highly desirable. The classical way of accessing XML data is through issuing structured queries, such as XPath/XQuery. However, in many applications it is inconvenient or impossible for users to learn these query languages. Besides, the requirement that the user needs to comprehend data schemas may be overwhelming or infeasible, as the schemas are often complex, fast-evolving or unavailable. A natural question to ask is whether we can empower users to effectively access XML data simply using keyword queries. Unlike text document search where the retrieval unit is an entire document, keyword search on XML data has every XML node as a retrievable unit, thus has a significant potential for fine-grained and high-quality results. Yet it poses a lot of unique challenges of inferring relevant fragments in XML data, composing query results, relevance based ranking, and result presentation. To gracefully process keyword searches on XML, an XML search engine should ideally satisfy a set of desiderata, varying from generating meaningful results to helping users quickly select relevant results. The desiderata include but are not limited to the following ones. Identifying Explicit Relevant Nodes. A user can specify the required information explicitly using keywords. However, not all nodes matching keywords are necessarily relevant to the query, which need to be distinguished by a search engine. Consider query "Galleria, Houston" on Figure 1 . For keyword "Houston", the match associated with city (6) is relevant, as it belongs to a store whose name is Galleria. The Houston node associated