Grant: Discovery Grants Program - Individual

Semi-structured and structured data, which include a significant portion of the world's enterprise and social data, are commonly stored in various forms such as RDBMSs, knowledge graphs, and social networks. These data repositories can either be classified as graphs or be transformed into graph structures. These graphs often possess node attributes, and each node is accompanied by descriptive text. Hence, these graphs are referred to as attributed graphs. While storing and manipulating graphs have seen notable advancements in the past, the progress in search and recommendation capabilities over graphs has been limited. As scientists and users seek valuable insights and actionable findings from their datasets, the demand for efficient capabilities becomes crucial. This research program aims to enable users to effectively explore graph databases and find relevant information. We will tackle significant problems, challenges, and opportunities associated with enhancing search, knowledge exploration, and recommendation in graphs. First, we investigate the problem of finding relevant answers in search over heterogeneous, incomplete, and uncertain graphs. Real graphs (e.g., social networks or biological networks) are heterogeneous, and could be incomplete and uncertain. Neural approaches have been shown effective in addressing related problems and we aim to tackle this problem using neural approaches. Second, we work on the problem of search in dual neural knowledge graphs (KGs). With recent advancement in large language models (LLMs), a new generation of KGs is being developed and is called dual neural KGs. In this generation, knowledge could be encoded explicitly as triples (the same as previous generations) or implicitly as embeddings. We study the problem of keyword search over KGs and explore how the capabilities of the dual neural KGs can be used in enhancing the search capabilities. Third, we propose novel algorithms for addressing distance queries in weighted stream graphs. Distance queries find the shortest distance (and path) between any two nodes in graphs. These queries are frequently used in search and recommendation operations in attributed graphs and can even be used to create or enhance graph neural representations. We plan to use a graph indexing method named 2-hop cover index and investigate how graph partitioning can be used to build the index. The outcomes of this research proposal have significant value for both Canadian and international businesses, as well as government institutions. We provide comprehensive training to students specializing in the fields of data management and data mining, positioning them favorably for academic and industrial job opportunities. Beyond the scientific impact of this research proposal, I look forward to having a positive impact on equity, diversity, and inclusion (EDI). I am committed to adopt inclusive research practices and offer mentorship and support to underrepresented students.