Constrained BERT BiLSTM CRF for understanding multi-sentence entity-seeking questions

Abstract

We present the novel task of understanding multi-sentence entity-seeking questions (MSEQs), that is, the questions that may be expressed in multiple sentences, and that expect one or more entities as an answer. We formulate the problem of understanding MSEQs as a semantic labeling task over an open representation that makes minimal assumptions about schema or ontology-specific semantic vocabulary. At the core of our model, we use a BiLSTM (bidirectional LSTM) conditional random field (CRF), and to overcome the challenges of operating with low training data, we supplement it by using BERT embeddings, hand-designed features, as well as hard and soft constraints spanning multiple sentences. We find that this results in a 12–15 points gain over a vanilla BiLSTM CRF. We demonstrate the strengths of our work using the novel task of answering real-world entity-seeking questions from the tourism domain. The use of our labels helps answer 36% more questions with 35% more (relative) accuracy as compared to baselines. We also demonstrate how our framework can rapidly enable the parsing of MSEQs in an entirely new domain with small amounts of training data and little change in the semantic representation.

“Opinion Target: Opinion Word” Pairs Extraction Based on CRF

“…(1) Frequency-based methods [1,2]; (2) Rule/template-based methods [3,4]; (3) Graph theory-based methods [6][7][8]; (4) Based on CRF or combined CRF and deep learning methods [5,[9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25].…”

“…Hu [14] designed a bidirectional long short-term memory (BiLSTM)-CRF model to achieve opinion extraction of product reviews. Danish [15] proposed a bidirectional encoder representations from transformers (BERT)-BiLSTM-CRF model for entity-seeking of multi-sentence. Huang [16] also used the BiLSTM-CRF model for sequence labeling tasks.…”

Joint Spatio-Textual Reasoning for Answering Tourism Questions

“…Our work builds on the recently-released POI entity-recommendation QA task (Contractor et al, 2019(Contractor et al, , 2020. Two approaches have been developed for this task: semantic parsing of unstructured user questions to query a semi-structured knowledge store (Contractor et al, 2020), and an end-to-end trainable neural model operating over a corpus of unstructured reviews to represent POIs. Neither of these approaches explicitly reason on spatial constraints, even though the questions contain them.…”

“…In order to get mentions of locations in questions, we manually label a set of 425 questions from the training set for location mentions. We then use a BERT-BiLSTM CRF (Contractor et al, 2020) based tagger trained on this set to label locations. We report some experiments using this model referred to as CSQA and compare it with CSRQA and spatio-textual CSRQA.…”

A conversation with Danish Contractor

“…Answers to these questions are usually names of entities such as the name of a restaurant. We are building methods that utilize large text collections of entity reviews to answer such questions automatically [5]. The work involves being able to understand a user's preference or constraints as well as distill knowledge from the vast collection of reviews describing each of those entities [6].…”