AGIF: An Adaptive Graph-Interactive Framework for Joint Multiple Intent Detection and Slot Filling

Abstract

In real-world scenarios, users usually have multiple intents in the same utterance. Unfortunately, most spoken language understanding (SLU) models either mainly focused on the single intent scenario, or simply incorporated an overall intent context vector for all tokens, ignoring the fine-grained multiple intents information integration for token-level slot prediction. In this paper, we propose an Adaptive Graph-Interactive Framework (AGIF) for joint multiple intent detection and slot filling, where we introduce an intent-slot graph interaction layer to model the strong correlation between the slot and intents. Such an interaction layer is applied to each token adaptively, which has the advantage to automatically extract the relevant intents information, making a fine-grained intent information integration for the token-level slot prediction. Experimental results on three multiintent datasets show that our framework obtains substantial improvement and achieves the state-of-the-art performance. In addition, our framework achieves new state-of-the-art performance on two single-intent datasets.

Smart Stacking of Deep Learning Models for Granular Joint Intent-Slot Extraction for Multi-Intent SLU

“…First of all, we change the way to annotate the data. [3], has extracted the relation between intents and slots, but the way to annotate the data do not serve our purpose to extract and map intents and slots at word/token level. For example, a sample taken from [3] 2 shows the following annotation (see TABLE I).…”

“…To solve this issue, we have re-annotated ATIS [2], and SNIPS [42] dataset. After annotation, we have applied a similar strategy (as applied by [3] to evaluate multi-intent system) to prepare MixATIS and MixSNIPS dataset.…”

“…This paper identifies the relation between slots and intents. Authors of [3], have proposed an Adaptive Graph-Interactive Framework (AGIF) for joint multiple intent detection and slot filling, where they introduce an intent-slot graph interaction layer to model the strong correlation between the slot and intents.…”

“…1 https://github.com/LeePleased/StackPropagation-SLU c) Identifying the relationship between intents and the corresponding slots. Most of the classifiers of this category (like - [1], and [3]), consider intent classification as a multi-level classification task and slot classification as a sequence labeling task that maps an input word sequence = ( 1 , 2 , … , ) onto the corresponding slot tags sequence = 1 , 2 , … . , (with the help of Begin/In/Out (BIO) format).…”

“…, (with the help of Begin/In/Out (BIO) format). Finally, [1], and [3], propose deep learning based interaction mechanism to combine intents and corresponding slots. To achieve these goals, they have used multitasking deep learning architectures.…”