Modeling and augmenting “searchable” online communities
MetadataShow full item record
In this dissertation, we have studied two distinct online communities, a question-answering (QA) system based on a social network and an information technology (IT) support organization. In a QA system, a query issued to search for an individual, a piece of information or some other resources needs to be forwarded to people who have knowledge about the destination of the target. Similarly, an IT support organization needs to quickly route a problem or incident reported by a customer to the appropriately-skilled experts. The primary objective of both systems is to ensure a timely resolution of the generated queries or incidents. Prior works in augmenting the search performances in the aforementioned communities have primarily focused on designing strategies that can achieve a high success rate in identifying the right person to forward a query such that, the query is handled by as few people as possible in order to be resolved. The strategies have assumed that a query can be processed by an individual, in a constant time, as soon as it has arrived. Such assumptions seem unrealistic since repeated selections of a subset of highly knowledgeable individuals will likely overload such individuals and lead to significant delays in responding and processing the pending questions or incidents. As our first contribution in this dissertation, we have modeled the communities as a queuing system in order to assess the performance of a search strategy in the presence of queuing delays. The model also incorporates the variation in the time taken by each user of the system to process a query, based on her depth and breadth of knowledge about the query. In order to ensure the authenticity of the derived theoretical models, we have modeled the characteristics of each critical component of the systems using both proven mathematical models and empirical results from analyses of real-world communities. Our second contribution is the introduction of novel search strategies for each community. The strategies take into account historical performances and interactions of community members to make more informed search forwarding decisions for future requests. We demonstrate that the overall performances of the systems can be improved significantly if the query forwarding and expert finding process is carefully designed to balance the query-load at the users in the system. We believe our results are a positive step towards building practical and effective QA system on social networks and managing incidents in an IT support system.