PPLRE Research Topics - Many-to-many Relations
Jump to navigation
Jump to search
Back to PPLRE Research Topics.
- Synopsis: Most Semantic Relation Algorithms are optimized for One-to-many Relation such as Organization/Location and Class/Subclass. For example, Snowball exploits the fact that patterns that associate a company with a different city must be rejected. The PPLRE task however involves a Many-to-many RElation, a Protein can be located in many Cellular Comparments, and vice versa. For example, the “LasA protease” protein of “P. aeruginosa” can be located in both the “cytoplasm” and the “outer membrane”. Ideas include … <tbd>
- Note: The problem might be alleviated if the training data contained all of the localizations for each protein in the training set. It would significantly reduce our training data examples if we asked our domain experts to provide only examples of proteins with all known localizations.
- Note: The problem would also be alleviated if the training data contained negative examples and the algorithm (e.g. LEILA) could make use of these examples.
- Note: One innate difficulty is to gather the correct distinguish during the Positive Sentence Harverting phase is a real relation from a misclassification. For example, in the Semantic Relation IsSiblingTo() the pattern [PERSON, brother of PERSON] will discover