CI Learner

Introduction

An increased exchange of scientific information across organizations and disciplines is one of the long-term goals of cyber-infrastructures and e-Science initiatives. In any such exchange of information, it is not difficult to identify one or more scientific communities responsible for the measurement, gathering and processing of scientific information. More challenging, however, is to understand the trust relations between members of these communities, whether the members are organizations or people. With a better understanding of trust relations, one may be able to compute trust recommendations for scientific information exchange, increasing in this way the acceptance of information by scientists. In this paper, we present CI-Learner, which is a systematic approach for extracting trust-related meta-information from scientific portals and related web sites. CI-Learner meta-information is organized as trust networks based on people, organizations, publications, and trust relations derived from publication co-authorships. Participation in a given trust network is restricted to organizations and people as identified by the CI-Learner information extraction process.

Approach

The CI-Learner approach extracts meta-information (e.g. names, urls) from scientific portals that present information about a scientific community. Scientist often express collaboration efforts in  producing scientific publications publishing their findings and results after processing information (e.g. sensor readings, experiment results). The creation of a scientific publication is possible due to the degree of trust co-authors of a publication deposit among themselves based on many criteria. Meaning that scientists tend to believe a publication if they have a degree of trust on the authors of the publication. We are interested on capturing and understanding the trust relations between these scientist.

Our approach, therefore, uses the PML ontology which captures a key relationship for our approach. A source, is either an agent or a document. People and organizations are instances of an agent, whereas a publication is an instance of a document. The relationship between an agent and document concept is that an agent is the creator of a document. Agents trust other agents by placing a degree of trust, where as agents believe the contents of a publication. Therefore, an agent believes a publication if the agent trusts the authors of the publication. 

The CI-Learner approach then extracts meta-information to build a social network of agents (e.g. person or organization) that are tied by relationships such as sharing cauthorship of a  document (e.g. publication). The CI-Learner approach makes use of Information Extraction and Information Retrieval techniques (e.g. wrappers, crawlers, metasearch engines) to extract meta-information for sources of interest.

Our approach starts at a scientific community portal extracting affiliated organizations. Next, we visit each organization looking for faculty and experts on the field of discourse. Finally, we query Google Scholar for publications of each of the people we were able to find during the extraction process. Co-authorship is captured from publications obtained and serve us as the trust relations conecting the members of the trust network. Finally, we make use of the Eigentrust algorithm to compute the aggregate trust values for each member of the network. These trust values represent the degree of trust for a  specific person as perceived by the whole trust network members and therefore used in our ranking.

Trust Networks

Gravity Network

This network presents members of the Gravity community (comming soon)

IRIS Network

This page presents the network for the members of the Incorporated Research Institutions for Seismology (IRIS) which dedicate their efforts to studies of Earth's interiors.