Y3 Milestone: Ry: Implement Edge Inference using Question Augmentation
richakanwar13 opened this issue · 0 comments
Description: We have shown that, given an edge to predict, there are small graph queries that are capable of returning good answers. However, the management of a large number of such queries, their use in a rational search strategy represents a new challenge. With a small number of relevant queries, these queries could be serially sent to Strider, and the results returned and collectively scored. With many such questions, and with many of those questions partially overlapping, we will investigate a more holistic approach in which many questions are sent to Strider and simultaneously explored. A related challenge is use of node-similarity in Strider’s searches, replacing the initial unwieldy attempt to pre-construct all possible questions involving similarity. Every node in an answer becomes an opportunity to apply a similarity edge, and it becomes important to integrate similarity deeply into Strider, rather than as a pre-processing step. A third enhancement to the ARAGORN pipeline will be the addition of scoring that takes into account the precision of an augmented query. Intuitively, answers to more precise queries have a higher chance of matching a user’s expectation, so we will investigate scoring methods taking this into account. Finally, we will continue algorithmic development, improving our ability to discover queries that imply previously unknown edges, and applying those algorithms to our data.