Representation learning for knowledge graphs (KGs) has focused on the problem\nof answering simple link prediction queries. In this work we address the more\nambitious challenge of predicting the answers of conjunctive queries with\nmultiple missing entities. We propose Bi-Directional Query Embedding (BIQE), a\nmethod that embeds conjunctive queries with models based on bi-directional\nattention mechanisms. Contrary to prior work, bidirectional self-attention can\ncapture interactions among all the elements of a query graph. We introduce a\nnew dataset for predicting the answer of conjunctive query and conduct\nexperiments that show BIQE significantly outperforming state of the art\nbaselines.\n