pilot news

Knowledge acquisition through crowdsourcing

December 5, 2017

Curious Cat uses conversation to gather information

As the market for high-tech mobile services becomes ever more crowded to accommodate an ever-growing number of mobile device users, context-aware services are growing in value. A ‘context-aware service’ is one that can provide useful information based on a user’s present location and current activity. If, for example, you are in Vienna looking for a good place to enjoy a reasonably priced coffee, a good mobile application can use your GPS coordinates to recommend one or more cafes within walking distance. But the providing of such seemingly basic information is far more complicated than one might think. The service must have a way of knowing what to recommend, and given answers will depend on input from the user—but the interaction between human intelligence and artificial intelligence is exceedingly complex, and is a vast area of ongoing research.

A six-person team (mostly Ljubljana-based) has published a recent study titled “Curious Cat—Mobile, Context-Aware Conversations: Crowdsourcing Knowledge Acquisition”. “To test a proposed novel approach [to the knowledge acquisition, or KA, process],” the authors write, “we built a proof-of-concept conversational assistant named Curious Cat, whose task is to provide its users with some interesting or useful information about the places they visit, while also being able to support incidental conversation ranging over common sense topics.”

The authors claim that the main scientific contribution of their paper is the presentation of a new approach to tackling the KA problem: “[Our] approach combines natural language crowdsourcing, usage of prior knowledge, context and newly acquired knowledge.” By being able to collect good-quality, verifiable knowledge, the authors believe that it will now be possible to “solve some higher-level tasks previously not possible.”

One of the novelties of Curious Cat is that it builds on the top of existing and newly acquired knowledge simultaneously, which facilitates an understanding of the user’s context. Increased use of the system automatically drives the KA process deeper and deeper by identifying missing pieces of knowledge and obtaining these missing elements through directed crowdsourcing and reasoning.

An equally important feature of the system, the authors argue, is that its novel crowdsourcing approach preserves user privacy. At the same time, it can store the user’s “beliefs about the world. This means that the knowledge the users provide through their answers can be local to the user and affect the other users only if promoted to general knowledge. If a user deliberately or accidentally misleads the KA process, this generally affects only the way the system interacts with that user, while having a minimal impact on the other users.”

Curious Cat formulates its questions directly to users, and the questions can take many forms. New information is only sent to other users as ‘new knowledge’ after it is validated. Additionally, the system can apply specific context to target those users with the strongest likelihood of being able to answer a question.

“The central and most important piece of contextual knowledge in our approach,” the authors continue, “is the user’s location and the duration of stay at this location. To facilitate this function, we use an improved implementation of the stay-point detection algorithm, which is able to cluster raw GPS coordinates and detect when a user is moving or staying at a particular location.”

Curious Cat has been publicly available online since the end of 2012, and is still running after several years of continuous activity. The system has more than 700 registered users, who have provided nearly 58,000 answers, which, combined with inferential possibilities, have generated more than 394,000 separate units of knowledge.

Click here to read the full paper.