-
MSc thesis: Discovering Higher Order Relations from Biomedical Text
-
Abstract: A discourse relation is a coherence relation that connects discourse segments expressing abstract
objects, i.e., events, facts, states or propositions. A biomedical relation, on the other
hand, exhibits a relationship between biomedical entities. When the abstract objects involved
in a discourse relation in a biomedical text correspond to biomedical relations (or biomedical
events, facts, etc.), we can infer a higher order relationship between those biomedical relations.
In this thesis, our goal is to extract such higher order relations from biomedical research articles.
These higher order relations can be used for question answering, knowledge discovery or
understanding reasoning in biomedical text. We have developed systems for parsing explicit
discourse relations and extracting biomedical relations from biomedical research articles. We
have evaluated these systems using public benchmark corpora and obtained promising results.
Finally, we have presented an algorithm that can extract higher order relations leveraging the
discourse relation parser and the relation extractor
-
Discourse Parser
-
A full discourse parser that extracts explicit discourse relations from English text. It has models trained on both the PDTB and the Biomedical Discourse Relation Bank. So, it can be used for parsing discourse relations from biomedical text as well as general English text.
-
Consider the following text: John will not come because he is sick. However, he may come tomorrow. So, we are going to talk about it tomorrow.
Sent-0: John will not come because he is sick .
Conn: because [4,4]
Sense: Contingency
Arg2Head: is 6
Arg1Head: will [0,1]
Sent-1: However , he may come tomorrow .
Conn: However [0,0]
Sense: Comparison
Arg2Head: may 3
Arg1Head: will [0,1]
Sent-2: So , we are going to talk about it tomorrow .
Conn: So [0,0]
Sense: Contingency
Arg2Head: going 4
Arg1Head: may [1,3]
The output follows the following format:
Sent-<Sentence#>: <Token 0> SPACE <Token 1> ... <Token (n-1)>
Conn: <Connective> TAB [<Start Token#>,<End Token#>]
Sense: <Sense>
Arg2Head: <Arg2 Head Token> TAB <Arg2 Head Token#>
Arg1Head: <Arg1 Head Token> TAB [<Arg1 Sentence#>,<Arg1 Head Token#>]
NewLine
-
Rule-based and ML-based biomedical relation extraction system.
-
LibSVM-Java-Kernel
-
Answering Natural Language Biomedical Queries using Answerset Programming
-
A query answering system that can provide answers to biomedical queries presented in
Attempto Controlled English.
The system converts a query into an answer set program and solves it using the
CLASP answer set solver.
For example, this system can give answers to queries like the following:
What are the drugs that treat a disease which causes Anxiety or which is related to HTR1A?
-
Automated Theorem Prover for First-Order Logic
-
A tableau-based automated theorem prover for first-order logic written in C++.
-
Bangla Speech Recognition System
-
A limited vocabulary, speaker independent, Bengali speech recognition system using CMU Sphinx speech framework.
-
Tautology Prover Applet
-
A tautology prover applet for propositional logic.
-
Generic Database Reader
-
An XML-configurable generic database reader in C++ that can read from MySQL, MS SQL Server, etc.
-
Model Builder for Propositional Partial Information Ionic Logic
-
Alt-LTC WebWatch
-
A Simpler Web Interface for LTC Web Watch.
|