A ranking pipeline for the shared touch Task 2: Argument Retrieval for Comparative Questions.
The goal of the second task was to rank documents of the CLueWeb12 corpus to support an imaginary user in decision making.ChatNoir was used as a baseline and ... was built around this search engine.
We provide two enviremonts.yml one for use with GPUs and one for use without GPUs. To use them Conda must be installed. Then the following command can be used to create the enviremont and to activate it.
conda env create -f environment.yml
conda activate touche
To return to the default environment
conda activate base
can be used.
The pipeline reloads all necessary and not locally available data. This may take some time depending on the available download rate. To use the pipeline, combiner.py must first be called with the flag -d to learn the parameters necessary for the SVM using a qrels file. All parts of the pipeline that are to be used later must be activated. It is possible to train several combinations one after the other and to use them later. All necessary data are stored locally under data.
python combiner.py [...] -d
After this, the pipeline can be used without the -d flag to generate a ranking.
usage: combiner.py [-h] [-p] [-e] [-w] [-m] [-a] [-b] [-u] [-t] [-v] [-d] [-o] [-s] Topics
positional arguments:
Topics File path to 'topics-task-2.xml'
optional arguments:
-h, --help show this help message and exit
-p, --Preprocessing Activate the Preprocessing (default: False)
-e, --QueryExpansion Activate the QueryExpansion (default: False)
-w , --WeightsMerging
Adding weights for merging responses
-m , --MergeMethod Method for merging responses (default: max)
-a, --Argumentative Activate the argumentative score (default: False)
-b, --Bert Activate the computation of a score via Bert (default: False)
-u , --Underscore Underscore for argument score (default: 0.55)
-t, --Trustworthiness
Activate the Trustworthiness score (default: False)
-v , --loglevel Set the detail of the log events (default: WARNING)
-d, --DryRun Start dry run to train the svm (default: False)
-o , --output File path where the output should be stored (default: ./out.trec)
-s , --size Size of the requested reply from ChatNoir (default: 100)