We examine the application of semantic context-aware Recommender Systems to improve interaction and navigation in a design-centric engineering domain. The small scale of this specialised environment renders most Web-scale solutions unsuitable, mandating tailored approaches. We report on initial work to identify challenges and promising categories of personalisation and adaptation together with relevant context features taken from the whole environment consisting of users, organisation, and documents to overcome the sparsity issue in professional Information Access.
In this paper, we examine social recommendation from an angle that has not yet received much attention. The angle is that of professional search and recommendation systems. In that case, social does not mean friends, followees and followers, or people who used the same Web site, but the social network of colleagues who work on similar projects or the same discipline. While this may limit the information to be gathered from the social circle, the focus on a speci c domain can partly make up for it. For example, an engineer in a product design company can have overlapping social circles and implicit connections. One can be the organisation chart of the hierarchy, setting her up in relation to people within her department, her supervisors, her team, or her sta . Additionally, she will be part of the engineers that work in the company in di erent departments and different elds, and she may also be part of one or multiple projects. She might additionally be part of a management group or a speci c specialization. All these roles and task mean di erent information access demands for her.
The setting we are examining is that of Knowledge-Based
Engineering (KBE), which is an approach used in
manufacturing and design engineering to not only capture available
process and product knowledge, but to use it systematically
in the design process. A focus lies on the reuse of
knowledge and knowledge sharing between the involved engineers
[
The KBE scenario is a good case for personalisation in professional search. The application domain involves knowledge workers and domain experts who need improved Information Access for complex problems and work tasks, based on complex and heterogeneous documents. Even if the knowledge base is a limited in-house system, the documents therein are mission-critical and contain valuable heterogeneous knowledge. Our aim is to enhance the Information Access to provide improved semantic navigation and interaction in information spaces.
We follow an exploratory approach to work towards
semantic contextual recommendations [
To understand and retrieve these relations, we need to
understand the context and tasks [
Usually, recommendations are generated by inferring
relations between documents based on users' interaction with
them, with common challenges regarding the level of
uncertainty for their results. In our case, a major challenge is
estimating which of the mentioned social circles contribute
to what extent to the search tasks and how they can be
made to work in her favour by supplying the right
information by direct recommendation, inform the ranking when
she is searching for information, of help her to better lter
and manage data, documents, or knowledge objects.
Another challenge is that personalisation goals can change a
lot during a work task, as she can take on di erent roles.
Finally, there is the problem of sparse and insu cient data,
as we have a much smaller number of documents, users, and
interactions than in large-scale systems, which can make
statistical approaches biased or wholly inapplicable [
We therefore exploit the rich domain-speci c context we
get from the scenario to better focus the recommendation.
In rst feasibility analyses, we take hints from the
literature [
The personalisation is still informed by the interaction of the users with the document database, but the context is used to o set the sparsity of interaction data. One exemplary task is to retrieve and understand the design decisions leading to a certain set of rules for structural components. The system will enable users to not only search for similar structures or access relevant documentation, but also for cases where similar problems had to be solved, which can give hints towards high-level alternatives. It might also be possible to learn certain work ows or best practices from other engineers. This is complicated by the fact that the more experienced an engineer is, the less they need to access related documents. This is one of the questions we aim to explore in interviews and later from the live system as part of the evaluation. 3.
We have presented our initial work towards the integration
of KBE and Recommendation Systems in a domain-speci c
and context-rich application scenario. The domain is
different from those examined in the literature which means
that we will have to heavily adapt and re ne existing
solutions as well as develop tailored methods. Our goal is to
use the described context information of the users and their
connections, the organisation, and the documents space to
enrich, support, and improve work ows in the
manufacturing engineering domain. We will further extend this
preliminary work towards a better understanding of the design
work ows and information needs [