| Name, Affiliation | |
|---|---|
| Editor: | Umutcan Şimşek, University of Innsbruck |
| Contributors: | Jürgen Umbrich, Onlim GmbH |
| Latest Version: | {docsify-updated} |
The concept of "browsing the web" is being replaced by automated agents that access to structured content. Dialogue systems are a natural way to interact with the machine readable data and services on the web. With DIALSWS, we will develop an approach for annotating web APIs with a lightweight semantic vocabulary based on schema.org and an approach for semi-automatically generating open and flexible goal-oriented dialogue systems based on those Web APIs.
Schema.org is one of the biggest collaborative vocabulary engineering projects and for the time being the de-facto standart. We aim to leverage the high-level adoption of the vocabulary by web content publishers an give them guidelines and tools to annotate not only their data but also their Web APIs. The semantic description of Web APIs, then will be used for semi-automated generation of intelligent agents, particularly dialogue systems.
The documentation below introduces the Unified Intent Modeling Ontology (UIMO) with some examples. For the details of Web API annotation with schema.org actions, see the WASA Specification.
There are many frameworks to help developers to create goal-oriented dialog systems. Such frameworks are built around the concept of intents, a piece of functionality that the dialog system supports, which is ultimately matched with the goal of the user. The fulfillment logic of those intents are typically strictly coupled with a specific framework and a backend service. This tight coupling creates a bottleneck in terms of interoperability and scalability. UIMO addresses this issue by providing a mechanism to describe an intent explicitly. Each intent is linked with its possible fulfillment methods that are represented with the WASA language.
TODO List reused ontologies and their purposes
?> You can find an overview of UIMO in the figure below. A human friendly documentation is coming soon. Until then please check out the vocabulary in Turtle format.
The core ontology already covers many aspects of dialog intents supported by various state-of-the art dialog system frameworks. Below we demonstrate how an intent modelled in UIMO can be mapped to a valid Alexa Intent:
A mapper can deterministically map UIMO types and properties to Amazon Alexa Interaction Model. The language model defines the supported intents and their slots, as well as the sample annotated utterances. For example, intents object in the interaction model is generated based on the intent URI and slot definitions. The utterances in the samples array are generated based on the text position annotations of :uttr1. The dialog model defines the interaction of the skill and the user based on the defined intents. The dialog.intents object defines the required slots with the elicitationRequired property for each slot. Each uimo:RequiredSlot is mapped to a slot definition with a true value for this property. The values of prompts are created based on the uimo:question values of uimo:RequiredSlot instances.
The intent generated based on a SearchAction with LodgingReservation result type is represent with UIMO as follows:
This intent is mapped to a DialogFlow intent:
Note that the intent generator contains a compabability layer that provides certain heuristics for framework specific mappings from UIMO.
The screenshots below show the intent generator tool.
An intent has fulfillments that are WASA Potential Actions
Slots are extracted from input and output descriptions of fulfillments
The training sentences can be annotated with the generated slots.
coming soon.
Şimşek, U., Kärle, E. & Fensel, D. (2018). Machine Readable Web APIs with Schema.org Action Annotations. Procedia Computer Science, 137, 255 - 261. doi: https://doi.org/10.1016/j.procs.2018.09.025
Simsek, U., Fensel, D.: Intent generation for goal-oriented dialogue systems based on schema.org annotations. In: The Prooceedings of 1st International Workshop on Chatbots co-located with ICWSM 2018 (2018), http://dblp.uni-trier.de/db/journals/corr/corr1807.html#abs-1807-01292 (Preprint)
This work is partially supported by:
This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License.
