End-to-End Entity Resolution and Question Answering Using Differentiable Knowledge Graphs

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4193–4200 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 4193 End-to-End Entity Resolution and Question Answering Using Differentiable Knowledge Graphs Armin Oliya Amir Saffari Priyanka Sen Tom Ayoola Amazon Alexa AI Cambridge, UK {aooliya,amsafari,sepriyan,tayoola}@amazon.com Abstract Recently, end-to-end (E2E) trained models for question answering over knowledge graphs (KGQA) have delivered promising results us- ing only a weakly supervised dataset. How- ever, these models are trained and evaluated in a setting where hand-annotated question enti- ties are supplied to the model, leaving the im- portant and non-trivial task of entity resolution (ER) outside the scope of E2E learning. In this work, we extend the boundaries of E2E learn- ing for KGQA to include the training of an ER component. Our model only needs the ques- tion text and the answer entities to train, and delivers a stand-alone QA model that does not require an additional ER component to be sup- plied during runtime. Our approach is fully dif- ferentiable, thanks to its reliance on a recent method for building differentiable KGs (Co- hen et al., 2020). We evaluate our E2E trained model on two public datasets and show that it comes close to baseline models that use hand- annotated entities. 1 Introduction The conventional approach for Question Answer- ing using a Knowledge Graph (KGQA) involves a set of loosely connected components; notably, an entity resolution component identifies entities men- tioned in the question, and a semantic parsing com- ponent produces a structured representation of the question. The programs resulting from combining these components can be executed on a knowledge graph (KG) engine to retrieve the answers. While this approach can be effective, collect- ing training datasets for individual components can be challenging (Dahl et al., 1994; Finegan-Dollak et al., 2018). For example, supervised seman- tic parsing requires training data pairing natural- language questions with structured queries, which is difficult to obtain. This has motivated many ef- forts in weakly supervised training (Chakraborty et al., 2021). Following recent breakthroughs in Figure 1: High-level architecture of the end-to-end model. One forward pass of RoBERTa extracts contex- tual embeddings for all components. Span Detection and Entity Resolution happen jointly to derive seed en- tities vector x0. The inference module performs multi- hop reasoning to reach answer entities vector ˆy machine translation (Bahdanau et al., 2015), a new goal is to directly optimize the entire chain of com- ponents end-to-end, without the need for interme- diate annotations. However, entity resolution (ER) is by and large a neglected component of E2E learning, and ex- isting weakly supervised solutions mostly assume question entities are either given or extracted by an external system. In practice, there’s a scarcity of quality training data for ER on questions, and poor entity extraction by out-of-domain models affects the overall performance of a KGQA system (Singh et al., 2020; Han et al., 2020). In this work, we present an end-to-end model for KGQA that learns to jointly perform entity res- …

Content truncated to 500 words