CS137 Project 3 Relation Extraction

Task

With the potential relation mention pairs provided, classify them into the correct relation.

File Description

• data: traning/dev/test instances, POS tagged sentences, constituency/dependency trees
• lib: maxent classifier and scripts
• lists: world knowledge lists (used for feature extraction)
• best_records: the best result trained from ./data/rel-train.gold and tested on ./data/rel-testset.raw
• dependency_tree.py: a data structure for dependency tree
• document.py: some data structures for document and instance representations
• feature.txt: feature configuration (currently the best feature combinations)
• features.py: all feature functions (for feature-based method)
• kernels.py: a pipeline for using scikit-learn SVM
• pipeline.py: a pipeline for using maxent
• tree_kernel.py: implementation of tree kernels (not successful)
• util.py: some utilities for loading data

Performance

(On tiara.cs.brandeis.edu)

Training: 286.60 sec

Decoding: 221.36 sec

P: 61.56 R: 34.74 F: 44.42

Usage

python pipeline.py --train data/rel-train.gold \
                   --test data/rel-test.raw \
                   --gold data/rel-test.gold \
                   --features feature.txt \
                   --task best_record

The pipeline will train on the train file, with the feature functions you specified with features argument, and test against the test file and evaluate its performance. All the results will be saved under the task folder.