The dataset download application link will be available for contestants after the registration for the competition. To ensure the privacy of game players, we have adopted an anonymization procedure. The dataset is composed of a training set, a testing set for each track, and an item description file. If you choose a reinforcement learning based algorithm, you need to construct your own training RL environment. See more details in the evaluation section.
\n\n\n\nThe trainset file is in text format, with space-delimiter and UTF-8 encoding. It involves more than 250k sessions, 400 items, and 40k users. The total size of the dataset is around 300MB. The columns of the trainset/testset files can be:
\n+ user_id: the unique identifier of the user\n+ user_click_history: the user's click history features. Data format: itemid1:timestamp1,itemid2:timestamp2,....\n+ user_protrait: the user's portrait features. Data format: feature1,feature2,....\n+ exposed_items: a nine-length vector composed of the unique identifier of the nine items\n+ labels: a nine-length 0/1 vector to indicate user purchase or not\n+ time: timestamp when the event happens\n\nFor Track I, the testset.csv has a zero-filled labels column, you need to predict it. For Track II, the testset.csv is consists of the users' information only, you need to generate the best nine-items combination for each user.
\n\n\n\nAnother item description file is also provided:
\n+ item_id: the unique identifier of item \n+ item_vec: the item's content features \n+ price: the item's price \n+ location: the n-th item list that item i is allowed to recommend\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Background", "content": "New Page: https://github.com/fuxiAIlab/RL4RS\n\nRL based RecSys is a data mining challenge organized in association with [IEEE BigData 2021 conference](http://bigdataieee.org/BigData2021/). The challenge is to promote applying reinforcement learning to industrial recommendation systems. The competition is sponsored by [FUXI AI Lab, Netease](https://fuxi.163.com/en/).\n\n**Special session at IEEE BigData 2021**: A special session devoted to the challenge will be held at the IEEE BigData 2021 conference. We will invite authors of selected challenge reports to extend them for publication in the conference proceedings (after reviews by Organizing Committee members) and presentation at the conference. The publications will be indexed in the same way as regular conference papers. The invited teams will be chosen based on their final rank, innovativeness of their approach, and quality of the submitted report. \n\nContact us: bigdatacup2021_rl4rs@service.netease.com", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "In 2020, retail e-commerce sales worldwide amounted to 5.23 trillion US dollars and e-retail revenues are projected to grow to 6.54 trillion US dollars in 2022. Such rapid growth promises a great future for the worldwide e-commerce industry signifying a strong market and increased customer demand. Besides the huge increment of the traffic volume, there has been a rapid growth of various recommendation scenarios, including bundle recommendation (a set of items), sequential item recommendation, and many others. It is worth exploring the various challenges that the modern e-commerce industry faces today. Most current e-commerce and retail companies build their recommender systems by implementing supervised learning based algorithms on their websites. The recommender systems are designed to maximize immediate user satisfaction in a greedy manner. However, the item-wise greedy recommendation strategy is imperfect fitting to the real recommendation system. With more and more new upcoming recommendation scenarios, more and more challenges have to be solved. For instance, in sequential recommendation scenarios, traditional methods often consider different ranking steps in a session to be independent and fail to maximize the expected accumulative utilities in a recommendation session. Moreover, in the news feed recommendation or bundle recommendation, the conversion rate of an item does not solely depend on itself. If an item is surrounded by similar but expensive items, the likelihood of buying it increases, known as the decoy effects. However, the possible combinations of all the items can be billions, which is an NP-hard problem and less explored in traditional supervised learning. To deal with these challenges, recent researches resort to adopting reinforcement learning for recommendations, in which the recommendation process is formulated as a sequential interaction between the user (environment) and the recommendation agent (RL agent). Reinforcement learning is a promising technology since the RL paradigm is inherently suitable for optimizing long-term user satisfaction directly, exploring the combinatorial space, and solving multi-step decision-making problems.\n\nThus, understanding how to build the RL-based recommendation system is practical and has a deep influence on the development of e-commerce platforms. To help tackle these challenges and further nourish the development in this field, we propose the following competing tasks, recall for reinforcement learning based recommender system challenge.\n\nSpecifically, we are motivated by one of the most popular games released by NetEase Games for providing an online item recommendation service. This item recommendation task is characterized by its mid-dimensional action spaces (roughly 400 item candidates after reducing) and special interaction rules. As is shown in Figure 1, for each user request, **the recommendation engine will respond with 3 item lists (3 items per list), and the next item list is locked until the items of the current list are sold out. The users' response depends on not only the current item but also the items of the next list.** This item recommendation problem can be naturally formalized as a multi-step decision-making problem in which the agent recommends an item or an item list in each step.\n\n**So, our goal is to recommend nine items for each user request to maximize the aggregated rewards.** Because other user satisfaction indicators are difficult to quantify, here we define the reward as the sum of the users' purchase amount of the nine items. An obvious baseline is to recommend the users' most likely items in each item list, but this strategy does not take into account the relationship between different item lists and cannot achieve the optimal solutions. We set up two tracks in this competition, see the 'Evaluation' Section.\n\n\n\n\n### Remarks:\n+ There are no overlap users between the test set and the train set. So the winning solutions need to avoid the over-fitting problem (reinforcement learning algorithms are prone to over-fitting).\n+ The training data is collected across three months and includes several sales campaigns (release new items). In Track II, the submission is evaluated by a pre-trained reinforcement learning environment. Simply speaking, the environment is a user purchase simulator trained by the data ten times the size of the training data. The test environment is only used to evaluate the submission and not accessible to participants. \n+ compare to the billions of possible combinations, the training data only covers limited state-action space. It means the RL algorithms may respond to unseen state-action confidently but falsely.\n+ We provide content features of the items and anonymized user features to help the participants explore solutions capable of handling the under-explored items well.\n\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "This is a page where you can include rules that participants must accept before joining. You may wish to include rules like:\n\n - Don't cheat!\n - Apply yourself!\n - Have fun!", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "### Track I : Item Combination Prediction\nThe task of Track I is to predict users' feedback to the given nine exposed items. It is not a simple point-wise supervised learning task. You need to give a prediction of 0/1 for each item and arrange the results as a nine-length space-delimited list (left to right top to bottom). The quality of submissions will be evaluated using the CategorizationAccuracy measure. Solutions will be evaluated in Kaggle and the results will be published on the public leaderboard. The score will be computed on a subset of the test dataset, fixed for all participants. The final evaluation will be performed after the completion of the competition using the remaining part of the test dataset. Those results will also be published online. It is important to note that only teams which submit a report describing their approach before the end of the challenge will qualify for the final evaluation.\n\n### Track II : Item Combination Generation (Recommended)\nThe task of Track II is to generate the best item combination (no repeat items) for each user to maximize the expected reward, i.e., the sum of the product of items' price and their purchase probability. It is a pure reinforcement learning task. The quality of submissions will be evaluated using the pure reinforcement learning measure, reward. You need to send your Kaggle 'User Name' as email title and submission.csv (1 Maximum Daily Submissions) as attachment to bigdatacup2021_rl4rs@service.netease.com . Solutions will be evaluated by a pre-trained large-scale RL environment. The results will be replied to immediately and published on the [Track II leaderboard](https://fuxi-up-research.gitbook.io/fuxi-up-challenges/challenge/bigdatacup2021-rl4rs-challenge). The score will be also computed on a subset of the test dataset, fixed for all participants. The final evaluation will be performed after the completion of the competition using the remaining part of the test dataset.\n\nFor every user in the test dataset, submission files should contain two columns: id and category (the predicted user purchase actions derived by your model). category should be a nine-length space-delimited list representing whether the users will purchase each items.
\nThe file should contain a header and have the following format:
\nid,category\n\n
1,1 1 1 0 1 0 0 0 0
For every user in the test dataset, submission files should contain two columns: id and itemids (the best item combination derived by your RL model). ItemIds should be a nine-length space-delimited list.
\nThe file should contain a header and have the following format:
\nid,itemids", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Prizes", "content": "
1,1 2 3 4 5 6 7 8 9
Track 1
\n1st Place - $1,000 + one free IEEE BigData 2021 conference registration
\n2nd Place - $500 + one free IEEE BigData 2021 conference registration
\n\nTrack 2
\n1st Place - $1,000 + one free IEEE BigData 2021 conference registration
\n2nd Place - $500 + one free IEEE BigData 2021 conference registration
\n\nThe award ceremony will take place during the special track at IEEE BigData 2021 conference (online).", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Code Requirement", "content": "Source code will also be required to be submitted, either through a publicly available repository on a Git-based version control hosting service such as GitHub for the final evaluation. All source codes are expected to be released as open-source software, utilizing some generally accepted licensing such as Apache License 2.0, GNU General Public License, MIT license, or others of similar acceptance by the Open Source Initiative.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Timeline", "content": "- August 31, 2021: deadline for submitting the solutions,\n- September 5, 2021: deadline for sending the reports, end of the competition,\n- September 12, 2021: online publication of the final results, sending invitations for submitting papers to the associated workshop at the IEEE Big Data 2021 conference,\n- October 10, 2021: deadline for submitting invited papers,\n- October 25, 2021: notification of paper acceptance,\n- November 15, 2021: camera-ready versions of accepted papers due.\n\nAll deadlines are at 11:59 PM UTC on the corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Organizers", "content": "Organizers:\nKai Wang, Netease FUXI AI Lab, wangkai02@corp.netease.com\nRunze Wu, Netease FUXI AI Lab, wurunze1@corp.netease.com\nZhene Zou, Netease FUXI AI Lab, zouzhene@corp.netease.com\nXudong Shen, Netease FUXI AI Lab, hzshenxudong@corp.netease.com\nTangjie Lv, Netease FUXI AI Lab, hzlvtangjie@corp.netease.com\nChangjie Fan, Netease FUXI AI Lab, fanchangjie@corp.netease.com\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }