Context

Relevant Wikipedia links:

• Context

In my currently running PhD is about context elicitation by user observation in the field of personal knowledge management.
The goal is to provide a context model which enables pro-active and unobtrusive support for a knowledge worker, i.e., there is an information assistant using the automatically generated user context. To be more concrete, my interest and contribution in context will have to worry about the following sub-topics:

• user context model: I am using RDF/S to represent the user's model. The context model uses common ontologies whenever possible, which is essential to infer rich potential assistance for the user at work. See [Schwarz2005] for more information.
• explanation: Whenever information is presented to the user, he should be able to look up, where the information came from and why it has been estimated to be relevant for him. This will not only lead to a more dependable system, but also enable meaningful user feedback.
• user observation: The user context is mainly fed automatically by applying user observation techniques. Observation plugins are installed into commonly used applications, such as Mozilla Thunderbird (e-mail) and Firefox (web browsing). The observed user actions are sent to the context elicitation framework.
• context elicitation: The observed user actions are the first contextual information we get from the user. The context elicitation framework will create contextual elements (CEs) from the observed so called Native OPerations (NOPs) of the user and enters the new CEs into the user context model. However, these CEs are only created at the lowest level of the whole pipelines elicitation architecture: Context elicitation modules will use lower-level CEs to estimate higher-level CEs.
  For instance, a recognized ViewEmail NOP will provide evidence for some relevant topic to be included in the context model. This happens because the text content of the viewed email is classified by a respective context elicitation module which knows the user's topic taxonomy (ontology).

Personal Knowledge Management

Relevant Wikipedia links:

• Personal Knowledge Management
• Organizational Knowledge Management

Eliciting the user's context to support his daily work has two implications:

1. If it works, the user will be able to get his work done better and faster because he gains time and stays in his flow. This is already a means of knowledge management.
2. The context elicitation and, hence, the context-sensitive support can only work if the system knows more about the user, his tasks, domains, and relations he draws between resources. Some of the information can be found already on his computer(s), e.g., in his file folders or in his email folders, other information has to be entered by the user directly. On the other side, the user does not want another source of distraction and additional work. So, he will only provide this information if the following two conditions are met:
   1. He himself will benefit most by adding this information.
   2. Entering this information costs but one or two clicks, i.e., does not cost time nor effort.
   If we manage to satisfy both conditions, the user will contribute to the organizational knowledge management, and he will do this freely which means, the quality of the entered information and meta-information will be high.

Human Computer Interaction (HCI)

Relevant Wikipedia links:

• Human Computer Interaction

The realization of a knowledge assistant is not a trivial task. On one hand, we have to make sure, that important and highly relevant information reaches the user as fast as possible, which means, we have to present it pro-actively and, maybe, together with some sort of alert. On the other side, especially these alert-style assistance, is what disturbs us computer users most. So, we have to balance pro-activeness with unobtrusiveness. Additionally we have to investigate how such information should be presented and how much information items he will be able to handle that was.

My research in HCI will not answer questions like the last one (how much items can the user handle), however, I will contribute to HCI by proposing and evaluating a context-sensitive assistance user interface.

Semantic Desktop

I am supporting and exploiting technologies and methodologies of the semantic desktop paradigm, as it coincides quite well with the modeling of a knowledge worker using his PC to get his work done. My research concerning modeling, maintainance, and retrieval of a user context model contributes to the semantic desktop research (which is part of the semantic web research).

Additionally, the realization of pro-active, context-sensitive assistant systems are important contributions to erect the semantic desktop. As these assistants utilize the (automatically) generated user context to come up with proposals or shortcuts of relevant information items (resources!) they allow a fast creation of relations and meta-data for resources at hand. Without such applications enabling easy and fast creation of RDF statements, the semantic web will never have a change to get running, because there is no data to work on. Hence, my small contribution will especially enable kick-starting the semantic web.

Case-Based Reasoning (CBR)

Relevant Wikipedia links:

• Case-Based Reasoning

My context elicitation framework applies case-based reasoning in several aspects:

• CBR methodology: Case-bases are a natural way of training/learning. I am storing user (inter)action patterns together with higher-level contextual elements (CEs) in a case base and use them to retrieve (estimate!) such higher-level CEs for similar user (inter)action patterns. Besides the retrieval, CBR provides intrinsic support for causal and provenance explanation
• Similarity measures are used in the context eliciation modules to compute the similarity or relevance of CEs. There is quasi no fuzzy reasoning and retrieval without similarity measures.
• CBR life-cycle: The estimated context is used to enable context-sensitive (and maybe pro-active) assistance for the user. If there is evidence about the user really accepting or denying the support the context elicitation can react according to that (implicit) user feedback. Hence, the context elicitation improves (boosts) after each user feedback.

Machine Learning

Relevant Wikipedia links:

• Machine Learning

I apply self-organizing feature maps (also known as Kohonen networks) whenever I need to visualize and exploit the similarity-based topology of objects (e.g., topics or documents).

Note: There is always a great overlap of research topics between several communities. As I am slightly inclined towards the CBR community, I am, for example, placing the similarity measures there.

Bayesian (Belief) Networks (BN, BBN)

The relationships of lower-level and higher-level contextual elements are modeled in a bayesian network style. The computation of their probabilities/confidences are done accordingly.

Semantic Web

Relevant Wikipedia links:

• Semantic Web

I am benefiting from research done in the semantic web concerning modeling, inferencing, and retrieval of contextual or context-related information. This holds especially for the distributed aspects of modeling, inferencing, and retrieval.

Semantic Desktop

As already mentioned I am contributing to some extend to the semantic desktop research. I am, of course, also heavily benefiting from the research done in this field.

Ontologies

Relevant Wikipedia links:

• Ontologies

Ontology research provides means of describing and infering information. I am specifying, as well as, incorporating ontologies to describe contextual or context-related information. I am appreciating research results regarding distributed inference techniques and methodologies.
Generally, I am supporting the religion of seeing ontologies as models of shared understandings of things.