This page has not been updated and has has information about some of my graduate life work.
(The cobot recommendation script is now taken offline.)
Try the Cobot Recommendation add-on for Openstudy conversations.
This script provides conversational recommendations for Openstudy.com conversations.
With the script, you can:
- Get web and conversational recommendations for an existing conversation.
- Get automatically notified for participating in other conversations based on your conversations, clicks and ratings.
Usage instructions: View screencast
Once you are setup to use cobot,
- Rate and Click on recommendations, cobot learns your preferences that way! (the more you use, the better it gets for you!)
- Fill in the survey to tell us how you feel, what you like and what you want to see!
Cobot - Conversational Information Access
Cobot is a community based agent recommender that provides socio-semantic recommendations in the conversation. Cobot extracts semantics from conversations, deciphers users' intent in conversations and recommends top documents (extracted from web search and previous pool of recommendations). Cobot also matches users from the community for this conversation and learns from it's recommendations.
Cobot does the following in the life-cycle of a conversation:
1. extracts semantics (concepts, categories and relations using ontology matching)
2. Deciphers conversational intent (Question, Acknowledgment, Advice, Disclosure, Reflection, Interpretation, Confirmation)
3. Formulates contextual queries
4. Retrieves document recommendations (using using a mixed bag of filtering methods)
5. Retrieves user recommendations (using using a mixed bag of filtering methods including spreading activation search on user models)
6. Checks engagement levels of recommended users for balancing recommendations (not deployed in the Openstudy script version)
7. Recommends users proactively (pushes instant notifications to recommended users)
8. Learns from ratings and clicks to improve recommendations socially
Cobot does the following in the life-cycle of a user in the community:
1. learns short term and long term models of the user.
2. registers feedback on entities such as users, documents and conversations for the collaborative filtering system
3. keeps track of users activity levels (not in the Openstudy script version)
4. provides general recommendations to the user based on her STM, LTM and social profiles (not in the Openstudy script version)
- semantic search
- Query re-formulation
- In conversation recommendation generation
- Social Search and matchmaking
- Real time recommendations
- Group based recommendations
- deciphering conversation intent (not just QnA) and prediction
- community balance and activity modeling for user engagement
- domain specific recommendations (Health, Education) and generic mode recommendations
- collaboration and interface
- Evaluation from the perspective of usability and user experience
STELLAR (SSTR Tools Enabling Lessons Learned Access and Reachback) (Summer 2009)
This project involved developing tools and methods for efficient reachback from lessons learned (LL)/best practices documents and repositories. The goal of the project was to facilitate commissioned forces to carry our Stability, Security, Transition and Reconstruction (SSTR) operations effectively with their limited human, experiential and knowledge based resources. We built a semantically annotated indexed Lessons Learned QnA Document Repository as part of the Georgia Tech sub-contract grant.
Medical Rule Extraction for Myocardial Perfusion Imaging literature using statistical, syntactic and semantic techniques.
iRemedi: Textual Case based Reasoning
We developed a Textual Case based Reasoning System that involves Graph based Knowledge Representation Techniques, Semantic Network creation and building PageRank and Spreading Activation based models for better retrieval and reasoning.
GeneTrek (Dr. Sham Navathe)
We explored techniques of Information Extraction and machine learning for Text mining of biomedical literature. Worked on feature extraction methods for classification of PubMed documents using supervised learning techniques.