![\"\"](\"https://s3.bmp.ovh/imgs/2021/09/be1737a20923996d.png\")
Congratulations to the Winners！
\nThanks for your participation. According to the public/private leaderboard and the completeness of submitted reports, the team ranking of each track is as follows:
\nTrack II：
\n1st Place
\nminhpham@usf.edu
\nMinhpham submits a highly completed paper. This paper proposes ComFOR algorithm, which significantly outperforms the state-of-the-art baselines including Deep Q-Network (DQN), Double Deep Q-Network(Double DQN), Batch Constrainted Deep Q-Learning (BCQ), and Conservative Q-Learning (CQL). Moreover, ComFOR can train significantly faster than other methods, roughly 10 times faster than the classic Q-Learning method and 30 times faster than the state-of-the-art BCQ.
\nhttps://github.com/minhhpham/IEEEBigData21_RL
2nd Place
\n18819253767@163.com
\nThis report proposes a simple and effective offline DQN based approach and it finally achieved second place in the competition. The DQN-based model is mainly based on the transformer layer and trained for more than 20 days.
\nhttps://github.com/chenhch8/rl4rs-dqn
3rd Place
\nmarkusjiafeng@163.com
\nThis report adopts an “Actor-Critic” RL scheme and tries a lot of reward shaping based on user history.
4th Place
\n3475635952@qq.com
\nBERT+Batch-Constrained Q-learning (CQL)
\nhttps://github.com/Ppaddington/BigData2021-Reinforcement-Learning-based-Recommender-System-Challenge
Track I：
\nWe are informed of the label leak phenomenon in Track I. Thank you, Yichao.
\n1st Place
\nlovers Team
\nzj140@mail.ustc.edu.cn
\nThey employ two base models, i.e., Attentive Interest Network and Hierarchical GRU Network, to learn the prediction task. Finally, they ensemble the predictions generated by the base models by the maximum voting principle.
\nhttps://github.com/zj140/RL4RS-lovers
2nd Place
\nlzhbrian Team
\nlzhbrian@gmail.com
\nThey use a two-headed transformer-based network to predict user feedbacks and unlocked sessions, along with the novel session-aware reweighted loss, multi-tasking with click behavior prediction, and randomness-in-session augmentation.
\nhttps://github.com/lzhbrian/bigdatacup2021.
3rd Place
\nTom&Jerry Team
\nwenlli0924@gmail.com
\nLightGBM
\nhttps://github.com/wenl-li/track1
4th Place
\nWuhu Fly Team
\nfeizhoucsu@foxmail.com
\nThis team joins both Track I and Track II, and achieves a good rank in each Track.
\nThey employ LSTM-based User simulator in Track I, and DQN/DDPG in Track II.
Congratulations to all the excellent participants!
\nQ: How about the award ceremony?
\nA: Congratulations! The award ceremony will take place during the special track at IEEE BigData 2021 conference (online).
Q: What is the associated workshop at the IEEE Big Data 2021 conference?
\nA:https://wi-lab.com/cyberchair/2021/bigdata21/scripts/ws_submit.php?subarea=SC
Q: Who will be invited papers to the associated workshop?\nA: The top 4 of either track will be invited. And the top 2 of either track must submit the relevant paper. These 8 papers have been accepted in principle, please ensure the quality. At least one authors of each paper need to register the IEEE Big Data conference (we will provide free registration for the top 2 teams of either track).
\nQ: The format of papers?
\nA: Papers should be formatted to IEEE Computer Society Proceedings Manuscript Formatting Guidelines (https://www.ieee.org/conferences/publishing/templates.html). Papers should be 4-10 pages (does not include references), in the IEEE 2-column format. The full instruction of format for the final camera-ready paper will be published at the conference in middle October.
Q： Following Timeline?
\nA：
\n (October 20, 2021): deadline for submitting invited papers,
\nOctober 25, 2021: notification of paper acceptance,
\nNovember 15, 2021: camera-ready versions of accepted papers due.
\n**All deadlines are at 11:59 PM UTC-12 on the corresponding day unless otherwise noted. **
Congratulations to the Winners！
\nThanks for your participation. According to the public/private leaderboard and the completeness of submitted reports, the team ranking of each track is as follows:
\nTrack II：
\n1st Place
\nminhpham@usf.edu
\nMinhpham submits a highly completed paper. This paper proposes ComFOR algorithm, which significantly outperforms the state-of-the-art baselines including Deep Q-Network (DQN), Double Deep Q-Network(Double DQN), Batch Constrainted Deep Q-Learning (BCQ), and Conservative Q-Learning (CQL). Moreover, ComFOR can train significantly faster than other methods, roughly 10 times faster than the classic Q-Learning method and 30 times faster than the state-of-the-art BCQ.
\nhttps://github.com/minhhpham/IEEEBigData21_RL
2nd Place
\n18819253767@163.com
\nThis report proposes a simple and effective offline DQN based approach and it finally achieved second place in the competition. The DQN-based model is mainly based on the transformer layer and trained for more than 20 days.
\nhttps://github.com/chenhch8/rl4rs-dqn
3rd Place
\nmarkusjiafeng@163.com
\nThis report adopts an “Actor-Critic” RL scheme and tries a lot of reward shaping based on user history.
4th Place
\n3475635952@qq.com
\nBERT+Batch-Constrained Q-learning (CQL)
\nhttps://github.com/Ppaddington/BigData2021-Reinforcement-Learning-based-Recommender-System-Challenge
Track I：
\nWe are informed of the label leak phenomenon in Track I. Thank you, Yichao.
\n1st Place
\nlovers Team
\nzj140@mail.ustc.edu.cn
\nThey employ two base models, i.e., Attentive Interest Network and Hierarchical GRU Network, to learn the prediction task. Finally, they ensemble the predictions generated by the base models by the maximum voting principle.
\nhttps://github.com/zj140/RL4RS-lovers
2nd Place
\nlzhbrian Team
\nlzhbrian@gmail.com
\nThey use a two-headed transformer-based network to predict user feedbacks and unlocked sessions, along with the novel session-aware reweighted loss, multi-tasking with click behavior prediction, and randomness-in-session augmentation.
\nhttps://github.com/lzhbrian/bigdatacup2021.
3rd Place
\nTom&Jerry Team
\nwenlli0924@gmail.com
\nLightGBM
\nhttps://github.com/wenl-li/track1
4th Place
\nWuhu Fly Team
\nfeizhoucsu@foxmail.com
\nThis team joins both Track I and Track II, and achieves a good rank in each Track.
\nThey employ LSTM-based User simulator in Track I, and DQN/DDPG in Track II.
Congratulations to all the excellent participants!
\nQ: How about the award ceremony?
\nA: Congratulations! The award ceremony will take place during the special track at IEEE BigData 2021 conference (online).
Q: What is the associated workshop at the IEEE Big Data 2021 conference?
\nA:https://wi-lab.com/cyberchair/2021/bigdata21/scripts/ws_submit.php?subarea=SC
Q: Who will be invited papers to the associated workshop?\nA: The top 4 of either track will be invited. And the top 2 of either track must submit the relevant paper. These 8 papers have been accepted in principle, please ensure the quality. At least one authors of each paper need to register the IEEE Big Data conference (we will provide free registration for the top 2 teams of either track).
\nQ: The format of papers?
\nA: Papers should be formatted to IEEE Computer Society Proceedings Manuscript Formatting Guidelines (https://www.ieee.org/conferences/publishing/templates.html). Papers should be 4-10 pages (does not include references), in the IEEE 2-column format. The full instruction of format for the final camera-ready paper will be published at the conference in middle October.
Q： Following Timeline?
\nA：
\n (October 20, 2021): deadline for submitting invited papers,
\nOctober 25, 2021: notification of paper acceptance,
\nNovember 15, 2021: camera-ready versions of accepted papers due.
\n**All deadlines are at 11:59 PM UTC-12 on the corresponding day unless otherwise noted. **
Congratulations Winners!
", "rawMarkdown": "Congratulations Winners!", "votes": null } ], "comments": [ { "id": 1512471, "author_name": "amalmohammed71215", "author_url": "", "post_date": "09/14/2021 09:20:57", "content": "Congratulations Winners!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1511181": "Congratulations to the Winners！\n\nThanks for your participation. According to the public/private leaderboard and the completeness of submitted reports, the team ranking of each track is as follows:\n\n**Track II：**\n1st Place\nminhpham@usf.edu\nMinhpham submits a highly completed paper. This paper proposes ComFOR algorithm, which significantly outperforms the state-of-the-art baselines including Deep Q-Network (DQN), Double Deep Q-Network(Double DQN), Batch Constrainted Deep Q-Learning (BCQ), and Conservative Q-Learning (CQL). Moreover, ComFOR can train significantly faster than other methods, roughly 10 times faster than the classic Q-Learning method and 30 times faster than the state-of-the-art BCQ.\nhttps://github.com/minhhpham/IEEEBigData21_RL\n\n2nd Place\n18819253767@163.com\nThis report proposes a simple and effective offline DQN based approach and it finally achieved second place in the competition. The DQN-based model is mainly based on the transformer layer and trained for more than 20 days.\nhttps://github.com/chenhch8/rl4rs-dqn\n\n3rd Place\nmarkusjiafeng@163.com\nThis report adopts an “Actor-Critic” RL scheme and tries a lot of reward shaping based on user history.\n\n4th Place\n3475635952@qq.com\nBERT+Batch-Constrained Q-learning (CQL)\nhttps://github.com/Ppaddington/BigData2021-Reinforcement-Learning-based-Recommender-System-Challenge\n\n**Track I：**\nWe are informed of the label leak phenomenon in Track I. Thank you, Yichao.\n1st Place\nlovers Team\nzj140@mail.ustc.edu.cn\nThey employ two base models, i.e., Attentive Interest Network and Hierarchical GRU Network, to learn the prediction task. Finally, they ensemble the predictions generated by the base models by the maximum voting principle.\nhttps://github.com/zj140/RL4RS-lovers\n\n2nd Place\nlzhbrian Team\nlzhbrian@gmail.com\nThey use a two-headed transformer-based network to predict user feedbacks and unlocked sessions, along with the novel session-aware reweighted loss, multi-tasking with click behavior prediction, and randomness-in-session augmentation.\nhttps://github.com/lzhbrian/bigdatacup2021.\n\n3rd Place\nTom&Jerry Team\nwenlli0924@gmail.com\nLightGBM\nhttps://github.com/wenl-li/track1\n\n4th Place\nWuhu Fly Team\nfeizhoucsu@foxmail.com\nThis team joins both Track I and Track II, and achieves a good rank in each Track.\nThey employ LSTM-based User simulator in Track I, and DQN/DDPG in Track II.\n\n\n\nCongratulations to all the excellent participants!\n\n\nQ: How about the award ceremony?\nA: Congratulations! The award ceremony will take place during the special track at IEEE BigData 2021 conference (online).\n\nQ: What is the associated workshop at the IEEE Big Data 2021 conference?\nA:https://wi-lab.com/cyberchair/2021/bigdata21/scripts/ws_submit.php?subarea=SC\n\n\n\n**Q: Who will be invited papers to the associated workshop?\nA: The top 4 of either track will be invited. And the top 2 of either track must submit the relevant paper. These 8 papers have been accepted in principle, please ensure the quality. At least one authors of each paper need to register the IEEE Big Data conference (we will provide free registration for the top 2 teams of either track).**\n\nQ: The format of papers?\nA: Papers should be formatted to IEEE Computer Society Proceedings Manuscript Formatting Guidelines (https://www.ieee.org/conferences/publishing/templates.html). Papers should be 4-10 pages (does not include references), in the IEEE 2-column format. The full instruction of format for the final camera-ready paper will be published at the conference in middle October.\n\nQ： Following Timeline?\nA：\n~~October 10, 2021~~ (October 20, 2021): deadline for submitting invited papers,\nOctober 25, 2021: notification of paper acceptance,\nNovember 15, 2021: camera-ready versions of accepted papers due.\n**All deadlines are at 11:59 PM UTC-12 on the corresponding day unless otherwise noted. **", "1512471": "Congratulations Winners!" }, "source": "meta" }