Benchmarking Natural Language Understanding Services for building Conversational Agents

TL;DR

The paper conducts a large-scale, cross-domain benchmark of four NLU services (Rasa, Watson, LUIS, Dialogflow) using a 25,716-utterance dataset spanning 21 domains, 64 intents and 54 entity types. It reveals that Watson excels at Intent classification but struggles with Entity precision, while LUIS, Dialogflow, and Rasa perform well on Entities, yielding similar overall F1 scores across platforms. The authors release both the dataset and evaluation toolkit to enable reproducibility and future benchmarking. Limitations include data noise, lack of dialogue context usage, and absence of ASR/spoken utterance evaluation, which they plan to address in future work.

Abstract

We have recently seen the emergence of several publicly available Natural Language Understanding (NLU) toolkits, which map user utterances to structured, but more abstract, Dialogue Act (DA) or Intent specifications, while making this process accessible to the lay developer. In this paper, we present the first wide coverage evaluation and comparison of some of the most popular NLU services, on a large, multi-domain (21 domains) dataset of 25K user utterances that we have collected and annotated with Intent and Entity Type specifications and which will be released as part of this submission. The results show that on Intent classification Watson significantly outperforms the other platforms, namely, Dialogflow, LUIS and Rasa; though these also perform well. Interestingly, on Entity Type recognition, Watson performs significantly worse due to its low Precision. Again, Dialogflow, LUIS and Rasa perform well on this task.