hybridRL_code_and_models

This repository contains code and models for this paper:  M. Ramezani , H.-C. Kum, G. Ilangovan, “Evaluation of Machine Learning Algorithms in a Human-Computer Hybrid Record Linkage System”, in the Association for the Advancement of Artificial Intelligence Symposium - Combining Machine Learning and Knowledge Engineering (AAAI-MAKE) 2021. (http://ceur-ws.org/Vol-2846/paper25.pdf) The presentation video: https://tube.switch.ch/videos/erw18iriu6

For further information contact Mahin: mahin@tamu.edu

Record Linkage

Record linkage refers to identifying the same entities across one or more databases when there is no unique identifier.

Hybrid Record Linkage

Our Hybrid Record Linkage framework combines the automatic and manual review process to achieve both scalability and high quality linkage results by allowing the automated algorithms to resolve majority of the linkages that have a high probability of being either a match or non-match, but also have the option to send ambiguous pairs to human experts for final determination to improve the linkage quality. You can use our trained models to conduct record linkage on your data or train a new model using your own dataset.

How to run the code

1. Prepare your pairs file (same format as NC_sample_raw_pairs.csv in the sample data folder)
2. Run code\1_feature_extraction.R to extract features from your pairs.
3. If you want to use n2v feature for first name and last name, run code\add_feature_name2vec.py on the output from step 2 and then run code\2_n2v_postprocess.R
4. Run code\3_model.R to train different ML RL models with your own data.
5. Run code\5__test_NC.R to test models on testing data.

Manual RL

Our code will save the uncertain pairs (that need manual review) as a CSV file. Each pair will be reviewed by two reviewer individually. If there is any disagreement, means one reviewer believes the pair is match and the other believes it is unmatch, two other reviewer will review that pair. If there is still disagreement, the four of them will have an open discussion meeting to resolve the pair.

You can find more information here: https://pinformatics.org/ppirl/mindfirl.php