Vision Statement

We believe that the availability of the Unstructured Information Management Architecture (UIMA), coupled with a powerful text indexing and processing server (Solr), which coupled with a particular approach to organizing the indexed resources (topic maps), all coupled with a rich domain of open source natural language processing (NLP) tools, makes it possible to craft a software agent framework with many of the capabilities of IBM's Watson. A blog post about the SolrSherlock project is here.

The project grows out of a much larger knowledge garden project, slides for which are found here.

Approach

We imagine our approach in terms of several steps to follow:

• Harvest information resources into a topic map
• Answer questions by navigating the topic map
• Study the map for latent connections

The opening step in approach is to inject a society of agents into the Solr ecosystem, directly in Solr's response handler chain. Here, we posit an agent-based architecture necessary to support many tasks. One of the tasks entail maintaining the topic map by way of a merge agent; another task is that of building the topic map by way of harvesting agents. 

Issues

Issues are legion; at the highest level they include but are not limited to:

1. Developer horsepower: the need for skilled, enthusiastic contributors
2. Funding: the project will entail ISPs for serving the platform, professional contributions on really complex algorithms and coding
3. Time: bandwidth to perform on the evolution of the platform

Solutions

Our solutions, for the SolrSherlock project, are emerging at Github in relation to the primary components of the agent-based architecture:

• SolrInjector
• SolrAgentCoordinator
• SolrAgentFramework
• SolrMergeAgent