I have completed my master's degree in the CS faculty under the supervision of Shaul Markovitch in 2009.
- My main research interest is Information Retrieval, and in particular Concept-Based IR.
I am following up work by Evgeniy Gabrilovich on Explicit Semantic Analysis (ESA) and researching its application to IR.
Our preliminary results were published in AAAI-08:
"Concept-Based Feature Generation and Selection for Information Retrieval"
and our full findings and analysis were published in the April 2011 issue of TOIS:
"Concept-Based Information Retrieval using Explicit Semantic Analysis"
- The main challenges in utilizing ESA in IR, are: a) unlike text categorization, the IR task does not include training data, and so noise in the generated features can cause more damage than improvement, and b) the power of our system lies mainly in generalization, whereas IR benchmarks seek to find needles in haystacks...
- We have also investigated the use of generated features for IR by using similarity metrics. The feature vectors generated by ESA can be used to compute semantic similarity between texts, and we have attempted to generalize classic tf.idf to compute relevancy by semantic similarity between query and documents.