Semantic Reasoning for the Networked Society
Ericsson Research has been selected to present at the 4th International Conference on Mobile Computing, Applications, and Services (MobiCASE2012) in Seattle, WA on October 11 & 12.
Ericsson’s vision of the Networked Society includes connecting devices, people and information, and to have devices work with people, for people and on behalf of people.
In order to achieve that, machines have to be able to derive high-level conclusions from low-level raw data sets and then to allow these machines to make further actions without any kind of human intervention. The results have to be produced quickly based on ever-changing and sometimes unreliable data sources. This was the challenge that we at Ericsson Research in San Jose and Tokyo teamed together to solve.
We developed a reasoning framework marrying the merits of ontology-based reasoning with other reasoning techniques. For those of you who don’t know, ontology is a semantic specification of a set of concepts and the relations between them. Ontology enables interoperability and supports sophisticated reasoning. Ontology-based reasoning is powerful, but it has well-known performance issues especially in cases where ontology is large and complicated.
During our research we learned that if you’re going to use semantic techniques for real-time reasoning, you need to optimize. This can be done using one or a combination of the following:
- optimize the size (and maybe the shape) of the ontology that describes the problem domain;
- use ontology only for the parts that really need it; and
- decouple ontology reasoning from the data sources.
Our reasoning framework employs two key mechanisms.
First, ontology is defined to include only high-level concepts. Without low-level concepts (such as measurements from sensors), an ontology significantly shrinks in size, which leads to improved reasoning performance. In addition, ontology is defined to include a conclusion tree. The conclusion tree (see below) contains possible conclusions at different levels of precision. In this way, it can provide the most accurate conclusion available at any given time.
To give some idea of what a conclusion tree looks like, let’s take the example of a surveillance system. The goal here is to detect the type of moving object based on sensors that identifies the speed of the object and the medium it moves in. See “Moving Object” in the figure below.
The second key mechanism is the decomposition of reasoning into two steps – “shallow reasoning” and “deep reasoning”.
Shallow reasoning takes the raw data from dynamic data sources and generates intermediate results. Deep reasoning takes the intermediate results from the shallow reasoning loaded into the ontology and draws more precise conclusions.
Working together, shallow reasoning handles the simple work, while deep reasoning makes efficient use of the powerful, but expensive semantic reasoning capability. Referring back to the example, the surveillance can determine first that the “Moving Object” is slow-moving, then that it is in the air and finally that it is in fact a “Flying Animal.” Details about this process can be found in our full report. View the presentation from the conference below.
Our next step is to apply the reasoning framework to interesting use cases in various domains. One example we are considering is in the Social Web of Things scenario, where devices talk to each other in a similar way that people interact on social networking sites. The reasoning framework could be applied to the smart home concept so that homeowners could connect to automated actions in their homes.
-- Bo Xing, Ericsson Research