Goal Recognition and Learning using Natural Language Processing within Smart Homes
The increase in use of smart devices have already changed the way people live, ranging from interactive smart homes to smart parking systems in cities. One of the main goal of WoT is to enable end users to exploit the diverse behaviors of these smart devices and be able to control/monitor and compose them to create new, added-value services. To reach such a goal, automated planning techniques are used as a potential approach. However, one of the main steps for such an automated planning system for IoT is to identify the goal/intention of the user.
There are many different personal assistants in the market to help end user’s reach their goal in ubiquitous environments. For instance, in Amazon Alexa the user’s intention is recognized by defining static rules and patterns using Amazon Alexa Skill, which is not a flexible mechanism. On the other hand, the Google assistant and Apple’s Siri have limited supports for IoT devices. When the user's speech can't be executed, the voice assistant will only return the results found by the search engine instead of learning the correct processing method from the user. This is because the current voice assistants are not designed specifically as IoT entrances, in more cases they act as search engine portals. However, for automated planning systems within the IoT domain it is vital that the user goal expressed using natural language is translated into a set of environmental parameters and device actions.
The objective of this Master thesis is to propose and implement a framework that is capable to convert natural language goals to machine executable language which takes into account the previously mentioned problems in the context of smart home domain. This application should understand direct goals as well as implicit goals expressed by the user.