A risk-aware conversational search system consisting of pretrained answer and question rerankers and a decision maker trained by reinforcement learning.

1. torch==1.4.0
2. transformers==3.4.0

1. Preprocess data.(alternatively, the processed datasets can be downloaded from datasets) Here we use MSDialog dataset as example. You can also set dataset_name to be 'UDC' for Ubuntu Dialog Corpus or 'opendialkg' for Opendialkg. First, download MSDialog-Complete.json into /data.

   $ cd data
   $ python3 data_processing.py --dataset_name MSDialog
   

   This will process and filter the data. All conversations that meet the filtering criterion are saved in MSDialog-Complete and will be automatically split into training and testing set. The others are save in MSDialog-Incomplete. The former is used for the main experiments and the latter is used for fine-tuning the rerankers only. The data processing code uses random.seed(2020) to fix the result of data generation.

2. Fine-tune pretrained reranker checkpoints on both the answer reranking and question reranking training samples (MSDialog as example). The training of the rerankers are based on ParlAI

   $ cd ParlAI
   $ python3 -u examples/train_model.py \
       --init-model zoo:pretrained_transformers/poly_model_huge_reddit/model \
       -t fromfile:parlaiformat --fromfile_datapath ../data/MSDialog-parlai-answer \
       --model transformer/polyencoder --batchsize 4 --eval-batchsize 100 \
       --warmup_updates 100 --lr-scheduler-patience 0 --lr-scheduler-decay 0.4 \
       -lr 5e-05 --data-parallel True --history-size 20 --label-truncate 72 \
       --text-truncate 360 --num-epochs 12.0 --max_train_time 200000 -veps 0.5 \
       -vme 8000 --validation-metric accuracy --validation-metric-mode max \
       --save-after-valid True --log_every_n_secs 20 --candidates batch --fp16 True \
       --dict-tokenizer bpe --dict-lower True --optimizer adamax --output-scaling 0.06 \
       --variant xlm --reduction-type mean --share-encoders False \
       --learn-positional-embeddings True --n-layers 12 --n-heads 12 --ffn-size 3072 \
       --attention-dropout 0.1 --relu-dropout 0.0 --dropout 0.1 --n-positions 1024 \
       --embedding-size 768 --activation gelu --embeddings-scale False --n-segments 2 \
       --learn-embeddings True --polyencoder-type codes --poly-n-codes 64 \
       --poly-attention-type basic --dict-endtoken __start__ \
       --model-file zoo:pretrained_transformers/model_poly/answer \
       --ignore-bad-candidates True  --eval-candidates batch
   

   $ python3 -u examples/train_model.py \
       --init-model zoo:pretrained_transformers/poly_model_huge_reddit/model \
       -t fromfile:parlaiformat --fromfile_datapath ../data/MSDialog-parlai-question \
       --model transformer/polyencoder --batchsize 4 --eval-batchsize 100 \
       --warmup_updates 100 --lr-scheduler-patience 0 --lr-scheduler-decay 0.4 \
       -lr 5e-05 --data-parallel True --history-size 20 --label-truncate 72 \
       --text-truncate 360 --num-epochs 12.0 --max_train_time 200000 -veps 0.5 \
       -vme 8000 --validation-metric accuracy --validation-metric-mode max \
       --save-after-valid True --log_every_n_secs 20 --candidates batch --fp16 True \
       --dict-tokenizer bpe --dict-lower True --optimizer adamax --output-scaling 0.06 \
       --variant xlm --reduction-type mean --share-encoders False \
       --learn-positional-embeddings True --n-layers 12 --n-heads 12 --ffn-size 3072 \
       --attention-dropout 0.1 --relu-dropout 0.0 --dropout 0.1 --n-positions 1024 \
       --embedding-size 768 --activation gelu --embeddings-scale False --n-segments 2 \
       --learn-embeddings True --polyencoder-type codes --poly-n-codes 64 \
       --poly-attention-type basic --dict-endtoken __start__ \
       --model-file zoo:pretrained_transformers/model_poly/question \
       --ignore-bad-candidates True  --eval-candidates batch
   

   This will download the poly-encoder checkpoints pretrained on reddit and fine-tune it on our preprocessed dataset. The fine-tuned model is save in ParlAI/data/models/pretrained_transformers/model_poly/.

   If you get an error of dictionary size mismatching, this is because that the pretrained model checkpoints has a dictionary that's larger than the fine-tune dataset. To solve this problem, before running the fine-tuning script, copy the downloaded pretrained dict file ParlAI/data/models/pretrained_transformers/poly_model_huge_reddit/model.dict to ParlAI/data/models/pretrained_transformers/model_poly/ and rename them to answer.dict. Then run the above fine-tuning script. Similar for the bi-encoder experiments.

   For bi-encoder fine-tuning, use the following command. When getting the dictionary size error, copy ParlAI/data/models/pretrained_transformers/bi_model_huge_reddit/model.dict to ParlAI/data/models/pretrained_transformers/model_bi/ and rename them to answer.dict.:

   $ cd ParlAI
   $ python3 -u examples/train_model.py \
       --init-model zoo:pretrained_transformers/bi_model_huge_reddit/model \
       -t fromfile:parlaiformat --fromfile_datapath ../data/MSDialog-parlai-answer \
       --model transformer/biencoder --batchsize 4 --eval-batchsize 100 \
       --warmup_updates 100 --lr-scheduler-patience 0 \
       --lr-scheduler-decay 0.4 -lr 5e-05 --data-parallel True \
       --history-size 20 --label-truncate 72 --text-truncate 360 \
       --num-epochs 12.0 --max_train_time 200000 -veps 0.5 -vme 8000 \
       --validation-metric accuracy --validation-metric-mode max \
       --save-after-valid True --log_every_n_secs 20 --candidates batch \
       --dict-tokenizer bpe --dict-lower True --optimizer adamax \
       --output-scaling 0.06 \
       --variant xlm --reduction-type mean --share-encoders False \
       --learn-positional-embeddings True --n-layers 12 --n-heads 12 \
       --ffn-size 3072 --attention-dropout 0.1 --relu-dropout 0.0 --dropout 0.1 \
       --n-positions 1024 --embedding-size 768 --activation gelu \
       --embeddings-scale False --n-segments 2 --learn-embeddings True \
       --share-word-embeddings False --dict-endtoken __start__ --fp16 True \
       --model-file zoo:pretrained_transformers/model_bi/answer\
       --ignore-bad-candidates True  --eval-candidates batch
   

   $ python3 -u examples/train_model.py \
       --init-model zoo:pretrained_transformers/bi_model_huge_reddit/model \
       -t fromfile:parlaiformat --fromfile_datapath ../data/MSDialog-parlai-question \
       --model transformer/biencoder --batchsize 4 --eval-batchsize 100 \
       --warmup_updates 100 --lr-scheduler-patience 0 \
       --lr-scheduler-decay 0.4 -lr 5e-05 --data-parallel True \
       --history-size 20 --label-truncate 72 --text-truncate 360 \
       --num-epochs 12.0 --max_train_time 200000 -veps 0.5 -vme 8000 \
       --validation-metric accuracy --validation-metric-mode max \
       --save-after-valid True --log_every_n_secs 20 --candidates batch \
       --dict-tokenizer bpe --dict-lower True --optimizer adamax \
       --output-scaling 0.06 \
       --variant xlm --reduction-type mean --share-encoders False \
       --learn-positional-embeddings True --n-layers 12 --n-heads 12 \
       --ffn-size 3072 --attention-dropout 0.1 --relu-dropout 0.0 --dropout 0.1 \
       --n-positions 1024 --embedding-size 768 --activation gelu \
       --embeddings-scale False --n-segments 2 --learn-embeddings True \
       --share-word-embeddings False --dict-endtoken __start__ --fp16 True \
       --model-file zoo:pretrained_transformers/model_bi/question\
       --ignore-bad-candidates True  --eval-candidates batch
   

   The fine-tuning code is based on ParlAI poly-encoder, but we modify several scripts for our needs. We do not recommended downloading the original ParlAI code and replace the ParlAI folder in this program. The original training of the encoders are done on 8 x GPU 32GB. We decrease the batch size and is able to run it on 4 x GPU 11GB (GeForce RTX 2080Ti).

3. Run the main experiments. To run the experiments, use the following code:

   $ python3  run_sampling.py --dataset_name MSDialog --reranker_name Poly --topn 1 --cv 0 > your_log_file
   

   --dataset_name can be 'MSDialog', 'UDC', or 'Opendialkg' currently. --reranker_name can be 'Poly' or 'Bi' currently. --topn means the top n reranked candidates are considered correct, i.e. --topn computes recall@1. The MSDialog dataset is too small, so it's recommended to run it using cross validation. When the dataset size is big enough or there is no need to run cross validation, simply use --cv -1 to turn off cross validation. The experiment would take a couple of hours to one day. So, it's recommended to save the results to a log file.

Please cite the following work if you use this code repository in your work:

@inproceedings{10.1145/3442381.3449893,
author = {Wang, Zhenduo and Ai, Qingyao},
title = {Controlling the Risk of Conversational Search via Reinforcement Learning},
year = {2021},
isbn = {9781450383127},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3442381.3449893},
doi = {10.1145/3442381.3449893},
pages = {1968–1977},
numpages = {10},
keywords = {conversational search, reinforcement learning},
location = {Ljubljana, Slovenia},
series = {WWW '21}
}

@article{10.1145/3507357,
author = {Wang, Zhenduo and Ai, Qingyao},
title = {Simulating and Modeling the Risk of Conversational Search},
year = {2022},
issue_date = {October 2022},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {40},
number = {4},
issn = {1046-8188},
url = {https://doi.org/10.1145/3507357},
doi = {10.1145/3507357},
month = {mar},
articleno = {85},
numpages = {33},
keywords = {risk control, reinforcement learning, Conversational search}
}