I have some confusions about figure1 and exposed_items:
\n@sword10086 If each time's recommendation contains nine items, Why did you say：\"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\" in Discription? It seems to be contradicted. Looking forward to your reply, Thank you!
", "rawMarkdown": "@sword10086 If each time's recommendation contains nine items, Why did you say：\"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\" in Discription? It seems to be contradicted. Looking forward to your reply, Thank you!", "votes": 2, "replies": [ { "id": 1364430, "postDate": "2021-06-25T01:53:16.663Z", "content": "e.g. The nine items can be derived by nine steps (think about the decoder of seq2seq), one item per step.
", "rawMarkdown": "e.g. The nine items can be derived by nine steps (think about the decoder of seq2seq), one item per step." }, { "id": 1374396, "postDate": "2021-07-03T09:16:53.953Z", "content": "But the second item lists are not unlocked until all items in the first item list are purchased, it seems like a combination optimization problem finding the best items combinations. The second items list should be showed to users when the first items list is available so that the recommendations in the second list would influence users' preference in the first item lists. It is nonsense to generate these 9 items in sequence since all 9 items are showed to users in one step.
", "rawMarkdown": "But the second item lists are not unlocked until all items in the first item list are purchased, it seems like a combination optimization problem finding the best items combinations. The second items list should be showed to users when the first items list is available so that the recommendations in the second list would influence users' preference in the first item lists. It is nonsense to generate these 9 items in sequence since all 9 items are showed to users in one step.", "votes": 3 }, { "id": 1374404, "postDate": "2021-07-03T09:23:51.873Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1376238, "postDate": "2021-07-05T01:43:35.423Z", "content": "Seq2seq is just an example. You can consider the 'unlock' influence in your custom model. The 'unlock' rule still holds when all 9 items are showed to users simultaneously. So it's still a combinatorial optimization problem.
", "rawMarkdown": "Seq2seq is just an example. You can consider the 'unlock' influence in your custom model. The 'unlock' rule still holds when all 9 items are showed to users simultaneously. So it's still a combinatorial optimization problem." }, { "id": 1376264, "postDate": "2021-07-05T02:19:47.183Z", "content": "If it's a combinatorial optimization, shouldn't a simulator be provided?
", "rawMarkdown": "If it's a combinatorial optimization, shouldn't a simulator be provided?" }, { "id": 1377640, "postDate": "2021-07-06T03:14:33.580Z", "content": "That's exactly the difficulty of applying RL into real applications. We need to build our own simulator from offline datasets.
", "rawMarkdown": "That's exactly the difficulty of applying RL into real applications. We need to build our own simulator from offline datasets.", "votes": 1 } ] }, { "id": 1360610, "postDate": "2021-06-22T08:08:30.313Z", "content": "I have some confusions about figure1 and exposed_items:
\n@sword10086 If each time's recommendation contains nine items, Why did you say：\"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\" in Discription? It seems to be contradicted. Looking forward to your reply, Thank you!
", "votes": 2, "replies": [ { "id": 1364430, "author_name": "sword10086", "author_url": "", "post_date": "2021-06-25T01:53:16.663000", "content": "e.g. The nine items can be derived by nine steps (think about the decoder of seq2seq), one item per step.
", "votes": 0, "replies": [] }, { "id": 1374396, "author_name": "BoscoTsang", "author_url": "", "post_date": "2021-07-03T09:16:53.953000", "content": "But the second item lists are not unlocked until all items in the first item list are purchased, it seems like a combination optimization problem finding the best items combinations. The second items list should be showed to users when the first items list is available so that the recommendations in the second list would influence users' preference in the first item lists. It is nonsense to generate these 9 items in sequence since all 9 items are showed to users in one step.
", "votes": 3, "replies": [] }, { "id": 1374404, "author_name": "", "author_url": "", "post_date": "2021-07-03T09:23:51.873000", "content": "", "votes": 0, "replies": [] }, { "id": 1376238, "author_name": "sword10086", "author_url": "", "post_date": "2021-07-05T01:43:35.423000", "content": "Seq2seq is just an example. You can consider the 'unlock' influence in your custom model. The 'unlock' rule still holds when all 9 items are showed to users simultaneously. So it's still a combinatorial optimization problem.
", "votes": 0, "replies": [] }, { "id": 1376264, "author_name": "BoscoTsang", "author_url": "", "post_date": "2021-07-05T02:19:47.183000", "content": "If it's a combinatorial optimization, shouldn't a simulator be provided?
", "votes": 0, "replies": [] }, { "id": 1377640, "author_name": "sword10086", "author_url": "", "post_date": "2021-07-06T03:14:33.580000", "content": "That's exactly the difficulty of applying RL into real applications. We need to build our own simulator from offline datasets.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1360617": "1. Nine items.\n2. Yes. He/She can see items in the second and third lists\n3. Recommend according to all features that describe the context.", "1360972": "@sword10086 If each time's recommendation contains nine items, Why did you say：\"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\" in Discription? It seems to be contradicted. Looking forward to your reply, Thank you!", "1360610": "I have some confusions about figure1 and exposed_items:\n1. For each time's recommendation, are there nine items or three items?\n2. If it's nine items, when user are choosing items in the first list, can he/she see items in the second and third lists?\n3. Are exposed_items recommended according to click_history or not?\n" } }