Going beyond sentences when applying tree kernels
Dmitry Ilvovsky · 2014
We go beyond the level of individual sentences applying parse tree kernels to paragraphs. We build a set of extended trees for a paragraph of text from the individual parse trees for sentences and learn short texts such as search results and social profile postings to take advantage of additional discourse-related information. Extension is based on coreferences and rhetoric structure relations between the phrases in different sentences. We evaluate our approach, tracking relevance classification improvement for multi-sentence search task. The search problem is formulated as classification of search results into the classes of relevant and irrelevant, learning from the Bing search results. We compare performances of individual sentence kernels with the ones for extended parse trees and show that adding discourse information to learning data helps to improve classification results.