PPLRE Research Topics - Document-based Analysis

Research Question

Most current Relation Recognition Algorithms discover Semantic Relations by treating the Corpus as a bag of Sentences (i.e. Sentence-level Analysis). Performance could be significantly improved by jointly analysing all of the sentences in an abstract (i.e. Discourse-level Analysis). Possible approaches include: 1) Perform Anaphora Resolution and Coreference Resolution on named entities and then analyze all of the sentences that include two or more of the sought Entity Types. This could improve Recall performance. 2) Look for evidence in the other sentences that helps the current methods by filtering out weak candidates. This could improve Precision performance.

E1) There are cases when the part of the relation is more clearly stated then the passage that contains all relations. See example below.
E2) Domain Experts themselves often cannot Classify with certainty whether a Sentence contains the Semantic Relation sought. Early into the project for example they requested the ability to read the whole Abstract to provide context to their Classification.

I1) Improve Recall by performing Anaphora Resolution and Coreference Resolution on named entities and then analyze sentences that include two or more of the sought Entity Types.
particularly Proteins which unite the two Binary Relations sought.

I2) Improve Precision by looking for evidence in the other sentences that helps the current methods by filtering out weak candidates.

See: PPLRE Corpus 610.a.0.