| Submitting Author | Affiliation | Title | Date Received | Decision | Notes |
|---|---|---|---|---|---|
| Léa Deleris | IBM | Automatically Constructing Bayesian Belief Networks using Text-mining Techniques | 7/27 | - | - |
| Jan Medlock | Clemson | Optimizing influenza vaccine policy | 8/5 | - | - |
| cell-content | cell-content | cell-content | cell-content | cell-content | cell-content |
| cell-content | cell-content | cell-content | cell-content | cell-content | cell-content |
| cell-content | cell-content | cell-content | cell-content | cell-content | cell-content |
Automatically Constructing Bayesian Belief Networks using Text-mining Techniques
Authors: Marcello Trovati, Léa Deleris, Jer Hayes, Luca Gasparini and Carlo Spaccasassi
Bayesian Belief networks (BBNs) are graphical models that capture independence relationships among random variables. They are often used as a modelling framework for risk management and medical decisions. The construction of a BBN can be done either through data, or when unavailable, through literature review or expert elicitation. While the first approach can be automated, the others require a significant amount of manual work which makes them impractical on a large scale.
Our research focuses on leveraging natural language processing (NLP) techniques to build BBNs from texts, focusing on the medical domain in our initial investigation. In our current implementation, a standard syntactic parser and sense taggers are used to create a basic network structure and novel analytics are then applied to create the BBN.
An essential aspect of text-information extraction for BBNs is the accurate identification of dependency (and independency) relations between concepts. This extends current research in NLP on relation extraction which has focused mostly on causation, a more specific concept.
We also seek to extract quantitative information related to the probabilities that underlie the BBN. This step implies determining the context into which the extracted information is embedded, to understand the conditioning scenario associated with the probabilities.
Finally, our approach includes the management of conflicting information, whether it deals with the structure of the network or with the probability statements. In particular, one needs to carefully elicit the source of the disagreement (in the text per se or as a result of the information extraction approach) and possibly evaluate the credibility of each statement.
Optimizing influenza vaccine policy
Authors: Jan Medlock, Clemson University, Department of Mathematical Sciences
The emergence of the 2009 H1N1 influenza A strain and delays in
production of vaccine against it illustrate the importance of
optimizing vaccine policy. Using a computational model, we determined
optimal vaccination policies with regard to multiple objective
functions: e.g. deaths, years of life lost, economic costs. We
explored the balance between vaccinating children who transmit most
and those individuals at greatest risk of mortality. In addition, we
have computed these optimal policies in the presence of uncertainty in
the epidemiology and costs of the emerging influenza strain. I will
discuss both the results and the general methods for optimizing this
policy.
