Memory Graph Networks for Explainable Memory-grounded Question Answering

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 728–736 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 728 Memory Graph Networks for Explainable Memory-grounded Question Answering Seungwhan Moon, Pararth Shah, Anuj Kumar, Rajen Subba Facebook Assistant {shanemoon, pararths, anujk, rasubba@}fb.com Abstract We introduce Episodic Memory QA, the task of answering personal user questions grounded on memory graph (MG), where episodic memories and related entity nodes are connected via relational edges. We cre- ate a new benchmark dataset first by gen- erating synthetic memory graphs with simu- lated attributes, and by composing 100K QA pairs for the generated MG with bootstrapped scripts. To address the unique challenges for the proposed task, we propose Memory Graph Networks (MGN), a novel extension of memory networks to enable dynamic expan- sion of memory slots through graph traver- sals, thus able to answer queries in which con- texts from multiple linked episodes and exter- nal knowledge are required. We then propose the Episodic Memory QA Net with multiple module networks to effectively handle various question types. Empirical results show im- provement over the QA baselines in top-k an- swer prediction accuracy in the proposed task. The proposed model also generates a graph walk path and attention vectors for each pre- dicted answer, providing a natural way to ex- plain its QA reasoning. 1 Introduction The task of question and answering (QA) has been extensively studied, where many of the existing applications and datasets have been focused on the fact retrieval task from a large-scale knowl- edge graph (KG) (Bordes et al., 2015), or machine reading comprehension (MRC) approaches given unstructured text (Rajpurkar et al., 2018). In this work, we introduce the new task and dataset for Episodic Memory QA, in which the model an- swers personal and retrospective questions based on memory graphs (MG), where each episodic memory and its related entities (e.g. knowledge graph (KG) entities, participants, ...) are repre- Figure 1: Illustration of Episodic Memory QA with user queries and memory graphs (MG) with knowledge graph (KG) entities. Relevant memory nodes are pro- vided as initial memory slots via graph search lookup. The Memory Graph Network walks from the initial nodes to attend to relevant contexts and expands the memory slots when necessary. The main QA model takes these graph traversal paths and expanded memory slots as input, and predicts correct answers via multiple module networks (e.g. COUNT, CHOOSE, etc.) sented as the nodes connected via corresponding edges (Figure 1). Examples of such queries in- clude “Where did we go after we had brunch with Jon?”, “How many times did I go to jazz concerts last year?”, etc. For Episodic Memory QA, a ma- chine has to understand the contexts of a question and navigate multiple MG episode nodes as well as KG nodes to gather comprehensive information to match the query requirement. While the ability of querying a personal database could lead to many potential applica- tions, previous work in this domain (Jiang et al., 2018) …

Content truncated to 500 words