Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Spot The Bot: A Robust and Efficient Framework for the Evaluation of Conversational Dialogue Systems

General Installation

You need to install a MongoDB v4.2.9 Server somewhere. All the conversation data is stored there. Unzip data_dump/MongoDump.zip and then import the files into your MongoDB (repeat this process for all 9 files):

mongoimport --db auto_judge_final --collection annotated-dialogues-full-convai2 --file annotated-dialogues-full-convai2.json --jsonArray --username <user_name>  --password <pw>

You need to install R...

You need to install Python 3.7, we suggest that you use Anaconda:

$ conda env create -f environment.yml

Adapt the config/annotation_app.json file as follows:

{
    "host": "ip_address of your MongoDB Server",
    "port": "port of mongodb",
    "user": "mongodb user name",
    "password": "pw of mognodb user",
    "database_name": "auto_judge_final",
    "package_collection_name": "packed-dialogues-full-{domain_name}",
    "sampled_collection_name": "sampled-dialogues-full-{domain_name}",
    "labelled_collection_name": "annotated-dialogues-full-{domain_name}",
    "local_port": 5003,
    "max_package_per_user": 3
}

Run the Annotation Tool

After you cloned the repository cd/autojudge_annotaiton:

To run the annotation tool:

$ python run.py

You can access the tool at localhost:5003

Ranking

After you cloned the repository cd/autojudge_annotaiton:

To get the Rankings based on Bootstrap Sampling (Table 1):

$ python templates\src\segment_analysis\segmented_bootstrap_sampling.py

To get the pairwise win rates (Table 1):

$ python templates\src\segment_analysis\win_function.py

To perform the stability experiment (Figure 3a):

$ python templates\src\segment_analysis\ranking_significance.py

To perform the leave-one-out experiment (Figure 3b):

$ python templates\src\segment_analysis\ranking_significance.py -lo 1

Survival Analysis

The survival analysis is implemented in R and uses the following packages:

To export the survival data from your annotations run python -m analysis.extract_event_data. This will create a csv file event_data.csv which is read by the R script.

Finally, run the R script at analysis/survival.R.

IAA

To run the label agreement analysis on e.g. the convai2 annotations, run

$ python analysis/inter_annotator_agreement.py sampled-dialogues-full-convai2.json

The annotations are stored in data_dump/MongoDump.zip

References

If you use this code, please cite us:

@inproceedings{deriu2020spot_the_bot,
  title = {{Spot The Bot: A Robust and Efficient Framework for the Evaluation of Conversational Dialogue Systems}},
  author = {Deriu, Jan and Tuggener, Don and von D{\"a}niken, Pius and Campos, Jon Ander and Rodrigo, Alvaro and, Belkacem, Thiziri and Soroa, Aitor and Agirre, Eneko and Cieliebak, Mark},
  booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing},
  address = {Online},
  year = {2020},
}

About

No description, website, or topics provided.

Resources

Releases

No releases published

Packages

No packages published