{"id":368,"date":"2024-01-10T18:01:26","date_gmt":"2024-01-10T09:01:26","guid":{"rendered":"https:\/\/jaeinlab.com\/?p=368"},"modified":"2024-01-11T18:43:52","modified_gmt":"2024-01-11T09:43:52","slug":"%ed%8e%98%ec%9d%b4%ed%8d%bc-%eb%a6%ac%eb%b7%b0-text-is-all-you-need-learning-language-representations-for-sequential-recommendation","status":"publish","type":"post","link":"https:\/\/jaeinlab.com\/?p=368","title":{"rendered":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0)"},"content":{"rendered":"\n
\n

Sequential Recommendation \ud0dc\uc2a4\ud06c\uc5d0 \uad00\uc2ec\uc744 \uac00\uc9c0\uace0 \uc788\ub358 \uc640\uc911, 2023\ub144\ub3c4 KDD\ub97c \uae30\uc6c3\uac70\ub9ac\ub2e4\uac00 \ubc1c\uacac\ud55c \ub17c\ubb38 “Text Is All You Need: Learning Language Representations for Sequential Recommendation” \ub9ac\ubdf0\uc785\ub2c8\ub2e4. \uac04\ub2e8\ud574 \ubcf4\uc774\ub294 idea\ub85c \uc0c8\ub85c\uc6b4 sequential recommendation paradigm\uc744 \uace0\uc548\ud55c \uc7ac\ubbf8\uc788\ub294 \ub17c\ubb38\uc774\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n

Keyword: Language Representation, Sequential recommendation, RecFormer, Transfer Learning<\/p>\n\n\n\n

<\/p>\n<\/blockquote>\n\n\n\n

Github \ub9c1\ud06c<\/a>\n\ub17c\ubb38 \ub9c1\ud06c (Arxiv)<\/a><\/p>\n<\/div>\n\n\n

\n
\"\"<\/figure>\n<\/div>\n\n\n
\ud55c \uc904 \uc694\uc57d: Item\uc758 key-value attribute pairs\ub97c \ud45c\ud604\ud55c Language representation\uc744 \ud65c\uc6a9\ud558\uc5ec, (\uc0c1\ub300\uc801\uc73c\ub85c) \uc81c\uc57d \uc5c6\ub294 transfer knowledge\ub97c \ud1b5\ud574 sequential recommendation \ud0dc\uc2a4\ud06c\uc5d0\uc11c SOTA performance\ub97c \ub0b8 RecFormer \uc544\ud0a4\ud14d\ucc98\ub97c \uc81c\uc548\ud55c \ub17c\ubb38.<\/code><\/pre>\n\n\n\n
Introduction<\/h2>\n\n\n\n
\uae30\uc874 Sequential Recommendation \ubc29\uc2dd\ub4e4\uc740 \uc77c\ubc18\uc801\uc73c\ub85c explicit\ud55c item IDs\ub098 \uc77c\ubc18\uc801\uc778 \ud14d\uc2a4\ud2b8 \ud53c\uccd0\ub97c \uc774\uc6a9\ud558\uc5ec sequence modeling\uc744 \uc9c4\ud589\ud558\uc600\ub294\ub370, \ud574\ub2f9 \ub17c\ubb38\uc5d0\uc11c\ub294 \uadf8\ub7f0 \ubc29\uc2dd\uc774 cold-start item \ubb38\uc81c\uc5d0 \ucde8\uc57d\ud558\uac70\ub098, \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc5d0 knowledge transferring \ud558\ub294 \uac83\uc5d0 \uc5b4\ub824\uc6c0\uc774 \uc788\ub2e4\uace0 \uc8fc\uc7a5\ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubb38\uc81c\ub4e4\uc744 \uc644\ud654\uc2dc\ud0a4\uae30 \uc704\ud574 \uc774 \uc5f0\uad6c\uc5d0\uc11c\ub294 user preference\uc640 item feature\uc744 language representation\uc73c\ub85c \ubaa8\ub378\ub9c1\ud558\ub294 \ubc29\ubc95 (RecFormer Framework)\uc744 \uace0\uc548\ud558\uc600\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n
\n
\"\"
\uadf8\ub9bc 3. \uae30\uc874 \ubc29\uc2dd\uc758 item ID sequence\uc640 RecFormer\uc5d0\uc11c \uc0ac\uc6a9\ub41c key-value attribute pair sequence\uc758 \ucc28\uc774<\/figcaption><\/figure>\n<\/div>\n\n\n
\uae30\uc874 \ubc29\uc2dd\uc758 item ID\ub85c \uad6c\uc131\ub41c sequence\uc640, RecFormer\uc5d0\uc11c \uc0ac\uc6a9\ud55c sequence\uc758 \ucc28\uc774\ub294 \uadf8\ub9bc 3\uc5d0\uc11c \ud655\uc778\ud558\uc2e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud655\uc2e4\ud788, item\uc758 title, brand, color\uc640 \uac19\uc740 \ub2e4\uc591\ud55c \uc815\ubcf4\ub85c \uad6c\uc131\ub41c sequence\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\ub2e4\uba74 \uc0c8\ub85c\uc6b4 item\uc5d0 \ub300\ud55c cold-start \ubb38\uc81c\uc5d0 \ub300\uc751\uc774 \uac00\ub2a5\ud560 \uac83 \uac19\uc2b5\ub2c8\ub2e4. \uadf8\ub7f0\ub370, \uc774\ub807\uac8c \ubcf5\uc7a1\ud558\uac8c \uad6c\uc131\ub41c key-value attribute pair \ub4e4\uc744 \ubaa8\ub378 input\uc73c\ub85c \uc0ac\uc6a9\ud558\uae30 \uc704\ud574\uc11c\ub294 \uc801\uc808\ud55c \ubcc0\ud615\uc774 \ud544\uc694\ud574 \ubcf4\uc785\ub2c8\ub2e4. <\/p>\n\n\n
\n
\"\"
\uadf8\ub9bc 4. Model input construction\uc5d0\uc11c\ub294 key-value attribute pairs\ub97c “\ubb38\uc7a5” \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658<\/figcaption><\/figure>\n<\/div>\n\n\n
\uadf8\ub9bc 4\ub97c \ubcf4\uc2dc\uba74, \ud574\ub2f9 \uc5f0\uad6c\uc5d0\uc11c\ub294 key-value attribute pair\uc744 “flatten”\ud558\uc5ec item \ubb38\uc7a5\uacfc \ube44\uc2b7\ud55c \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\ud558\ub294 \ubc29\ubc95\uc744 \ucc44\ud0dd\ud558\uc600\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c “\ubb38\uc7a5” \ud615\ud0dc\ub77c\uba74, \uae30\uc874\uc5d0 sequence\ub97c \ucc98\ub9ac\ud558\ub294 \ubaa8\ub378 input\uc73c\ub85c \uc0ac\uc6a9\ud558\uae30\uc5d0 \uc801\uc808\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n\n\n\n
RecFormer Model<\/h2>\n\n\n
\n
\"\"
\uadf8\ub9bc 5. RecFormer \uc804\uccb4 \uad6c\uc870<\/figcaption><\/figure>\n<\/div>\n\n\n
\uadf8\ub9bc 5 (a)\uc5d0\uc11c \ubcf4\uc774\ub294 \uac83\ucc98\ub7fc, RecFormer\uc740 multi-layer BERT \uad6c\uc131\uc744 \b\ud3ec\ud568\ud55c Longformer \ubaa8\uc591\uc758 \uad6c\uc870\ub85c \ub514\uc790\uc778 \ub418\uc5c8\uc2b5\ub2c8\ub2e4. \ubaa8\ub378 \uc778\ud48b\uc73c\ub85c\ub294 \uc704\uc5d0\uc11c flatten\ub418\uc5b4 \ub9cc\ub4e4\uc5b4\uc9c4 item “sentence”\uac00 \uc0ac\uc6a9\ub418\ub294\ub370, \ub2e4\uc74c\uacfc \uac19\uc774 \ud45c\uae30\ub429\ub2c8\ub2e4.<\/p>\n\n\n\n
$$X = \\{[CLS], T_n, T_{n-1}, …, T_1  \\}$$<\/p>\n\n\n\n
\uc5ec\uae30\uc5d0\uc11c \uac01 \\(T_i\\) \uc544\uc774\ud15c \\(i\\)\ub97c \ub098\ud0c0\ub0b4\ub294 “sentence” \uc785\ub2c8\ub2e4.<\/p>\n\n\n\n
Token Embedding<\/h3>\n\n\n\n
RecFormer\uc740 language understanding\uacfc sequence pattern \ubaa8\ub450\ub97c \uc774\ud574\ud558\uc5ec \ucd94\ucc9c \ud0dc\uc2a4\ud06c\ub97c \ud478\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4. \ub54c\ubb38\uc5d0 LM\uacfc SASRec (Self-attentive Sequential Recommenders)\ub97c \uacb0\ud569\ud558\ub294 \ubc29\ubc95\uc744 \uc0ac\uc6a9\ud588\uc2b5\ub2c8\ub2e4. \ucd1d \ub124 \uc885\ub958\uc758 embedding\uc774 \uc0ac\uc6a9\ub418\ub294\ub370\uc694. \uac01\uac01 Token embedding, Token position embedding, Token type embedding, Item position embedding \uc785\ub2c8\ub2e4.<\/p>\n\n\n\n
    \n
  • Token embedding \\(A_w\\): \uc774\uc804 Sequential Recommender\uc758 item embedding\uc744 \uc0ac\uc6a9\ud558\uc9c0 \uc54a\ub294 RecFormer\uc740 \uac70\uc758 Token embedding (word token embedding)\uc744 \ud1b5\ud574 interaction sequence\ub97c \uc774\ud574\ud55c\ub2e4\uace0 \uc0dd\uac01\ud574\uc8fc\uc2dc\uba74 \ub429\ub2c8\ub2e4. <\/li>\n\n\n\n
  • Token position embedding \\(B_w\\): LM\uc5d0\uc11c\uc640 \ube44\uc2b7\ud558\uac8c, Transformer\uac00 sequential pattern\uc744 \uc774\ud574\ud558\uae30 \uc704\ud55c position embedding \uc785\ub2c8\ub2e4.<\/li>\n\n\n\n
  • Token type embedding \\(C_w\\): token\uc774 \\(C_{[CLS]}, C_{Key}, C_{Value} \\in \\mathbb{R}^d\\) \uc911 \uc5b4\ub514\uc5d0 \uc18d\ud558\ub294\uc9c0 \uad6c\ubcc4\ud558\uae30 \uc704\ud55c \ubd80\ubd84\uc785\ub2c8\ub2e4. <\/li>\n\n\n\n
  • Item position embedding \\(D_w\\): sequence\uc5d0\uc11c item\uc758 \uc704\uce58\ub97c \uad6c\ubcc4\ud558\uae30 \uc704\ud55c \uc784\ubca0\ub529\uc785\ub2c8\ub2e4. Token position embedding\uc774 \ud1a0\ud070 \uac04 \uc21c\ucc28\uc801 \uad00\uacc4\ub97c \uc774\ud574\ud558\ub294 \ub370 \ud544\uc694\ud55c \ubc18\uba74, Item position embedding\uc740 \uc0c1\ud638\uc791\uc6a9 \uc2dc\ud000\uc2a4 \ub0b4\uc5d0\uc11c \uac01 \uc544\uc774\ud15c\uc758 \uc704\uce58 \uc815\ubcf4\ub97c \uc774\ud574\ud558\ub294 \ub370 \ud544\uc694\ud569\ub2c8\ub2e4 (\uc544\uc774\ud15c \uac04\uc758 \uc21c\ucc28\uc801 \ud328\ud134\uacfc \ub2e8\uc5b4 \ud1a0\ud070\uacfc \uc544\uc774\ud15c \uac04\uc758 \uc5f0\uad00\uc131 \uc774\ud574).<\/li>\n<\/ul>\n\n\n\n
    \ub2e4\uc74c\uc73c\ub85c, input sequence \\(X\\)\uc758 \ub2e8\uc5b4 \\(w\\)\ub294 \uc774 \ub124 \uac00\uc9c0 embedding\uc744 concat\ud558\uc5ec LayerNorm\uc744 \uc801\uc6a9\ud558\uc5ec \ucd5c\uc885 word embedding \\(E_w\\in\\mathbb{R}\\)\ub97c \uc5bb\uc2b5\ub2c8\ub2e4. \uc804\uccb4 sequence\ub294 \\(E_X = [E_{[CLS]}, E_{w_1}, …, E_{w_l}] \\in \\mathbb{R}^{(l+1) \\times{d}} \\) \ub85c \ub098\ud0c0\ub0bc \uc218 \uc788\uc73c\uba70, \\(l\\)\uc740 user’s interaction sequence\uc758 maximum \ud1a0\ud070 \uac2f\uc218\uc785\ub2c8\ub2e4.<\/p>\n\n\n\n
    Final Prediction<\/h3>\n\n\n\n
    \uc774\ub807\uac8c \ub9cc\ub4e4\uc5b4\uc9c4 embedding sequence\ub294 Longformer \ub808\uc774\uc5b4\ub97c \ud1b5\uacfc\ud558\uc5ec \ucd5c\uc885 \uc544\uc6c3\ud48b \\([h_{CLS}, h_{w_1}, …, h_{w_l}] \\)\uc744 \uc5bb\uac8c \ub418\uba70, BERT\uc758 classification\uacfc \ube44\uc2b7\ud558\uac8c \\(h_{[CLS]}\\)\uac00 item representation \\(h_i\\)\uac00 \ub429\ub2c8\ub2e4. <\/p>\n\n\n\n
    \ucd5c\uc885 \uc608\uce21\uc740 interaction sequence \\(s\\)\uc640 \uc544\uc774\ud15c \\(i\\)\uc758 cosine similarity \uacc4\uc0b0\uc73c\ub85c \uacc4\uc0b0\ub418\uc5b4, score\uac00 \uac00\uc7a5 \ub192\uc740 item\uc744 next item\uc73c\ub85c \ucd94\ucc9c\ud574\uc8fc\uac8c \ub429\ub2c8\ub2e4.<\/p>\n\n\n\n
    Learning Framework<\/h2>\n\n\n\n
    \ud6a8\uacfc\uc801\uc778 \ubaa8\ub378\ub9c1\uc744 \uc704\ud574, \ud574\ub2f9 \uc5f0\uad6c\uc5d0\uc11c\ub294 RecFormer\uc5d0 pre-training\uacfc two-stage finetuning\uc744 \ud3ec\ud568\uc2dc\ucf30\ub2e4\uace0 \ud569\ub2c8\ub2e4.<\/p>\n\n\n\n
    Pre-training<\/h3>\n\n\n\n
    \uae30\ubcf8\uc801\uc73c\ub85c pre-training \ubaa9\uc801\uc740 \ud0dc\uc2a4\ud06c\ub97c \uc880 \ub354 \uc798 \uc218\ud589\ud558\uae30 \uc704\ud55c \uc815\ubcf4\ub97c parameter initialization \uacfc\uc815\uc5d0 \uc5b4\ub290\uc815\ub3c4 \ubbf8\ub9ac \ub2f4\uace0\uc790 \ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774\ubc88 \uc5f0\uad6c\uc5d0\uc11c\ub294 language understanding\uacfc recommendation\uc774\ub77c\ub294 \ub450 \uac00\uc9c0 \ubaa9\ud45c\ub97c \ub2ec\uc131\ud558\uae30 \uc704\ud574, Masked Language Modeling (MLM)\uacfc item-item constrastive task\ub97c \ud65c\uc6a9\ud588\ub2e4\uace0 \ud569\ub2c8\ub2e4.<\/p>\n\n\n\n
    \uc800\uc790\ub294 MLM\uc744 \ud65c\uc6a9\ud558\uba74 general language corpus\uc640 item text \uac04\uc758 language domain gap \uc904\uc77c \uc218 \uc788\ub2e4\uace0 \uc774\uc57c\uae30\ud569\ub2c8\ub2e4. MLM \uac1c\ub150\uc740 \uc774\ubbf8 \uc798 \uc54c\ub824\uc838 \uc788\uae30\uc5d0, \uad6c\uccb4\uc801\uc778 \uc774\uc57c\uae30\ub294 \uc5ec\uae30\uc11c \uc2a4\ud0b5\ud569\ub2c8\ub2e4. MLM \uad00\ub828 \uc138\ud305 (e.g., random token probability)\uc740 \ub17c\ubb38\uc744 \ucc38\uc870\ud558\uc2dc\uba74 \ub418\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n
    item-item contrastive (IIC) \ud0dc\uc2a4\ud06c\ub294 next item prediction\uc5d0\uc11c \uc790\uc8fc \ud65c\uc6a9\ub418\ub294 \ubc29\ubc95\uc778\ub370, \uc27d\uac8c \ub9d0\ud574\uc11c negative sampling\uc744 \uae30\ubc18\uc73c\ub85c \ud55c contrastive learning\uc758 \uc77c\uc885\uc774\ub77c \uc0dd\uac01\ud558\uc154\ub3c4 \ubb34\ubc29\ud560 \uac83 \uac19\uc2b5\ub2c8\ub2e4. \ub208\uc5ec\uaca8\ubcfc \ub9cc\ud55c \uc810\uc740, negative sampling\uc744 \ud1b5\ud574 \uc9c1\uc811 negative instance\ub97c \uc0dd\uc131\ud55c \uac83\uc774 \uc544\ub2c8\ub77c, in-batch next items\ub97c negative instance\ub85c \ud65c\uc6a9\ud558\uc600\ub2e4\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774\ub807\uac8c \uc9c4\ud589\ud55c \uc774\uc720\ub294, \uae30\uc874 \ucd94\ucc9c\uc2dc\uc2a4\ud15c\uc5d0\uc11c \uac00\uc9c0\uace0 \uc788\ub294 item embedding table\uc774 RecFormer\uc740 \uc5c6\uae30 \ub54c\ubb38\uc5d0, batch\ub9c8\ub2e4 \uc544\uc774\ud15c\uc744 re-encode\ud560 \uc218\uac00 \uc5c6\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4. \uc774 \ubc29\ubc95\uc740 \uc9c1\uc811 sampling\ud55c \uac83\uc774 \uc544\ub2c8\ub77c, \ub2e8\uc21c\ud788 \uac19\uc740 batch \ub0b4 \ub2e4\uc74c item\uc744 negative\ub85c \uac00\uc815\ud574\ubc84\ub9ac\ub294 \uac83\uc774\uae30 \ub54c\ubb38\uc5d0 false negative\uac00 \ubc1c\uc0dd\ud560 \ud655\ub960\uc774 \uc788\uc73c\ub098, \uc800\uc790\ub294 pre-training dataset \uc0ac\uc774\uc988\ub97c \uace0\ub824\ud588\uc744 \ub54c \ubb38\uc81c\uac00 \ub420 \uc815\ub3c4\ub294 \uc544\ub2c8\ub77c\uace0 \uc8fc\uc7a5\ud569\ub2c8\ub2e4 (+ \uc5b4\ucc28\ud53c pre-training \ub2e8\uacc4\uc774\uae30 \ub54c\ubb38\uc5d0 \ucd5c\uc885\uc801\uc778 performance\uc5d0 \ub300\ud55c \uc601\ud5a5\uc740 \uadf9\ud788 \uc791\uae30\ub3c4 \ud569\ub2c8\ub2e4).<\/p>\n\n\n\n
    Pre-training stage\uc5d0\uc11c\ub294 \uc544\ub798 \uc218\uc2dd\uacfc \uac19\uc774, MLM\uacfc IIC task\ub97c multi-task training \ubc29\uc2dd\uc73c\ub85c jointly \ucd5c\uc801\ud654\ud569\ub2c8\ub2e4. \\(\\lambda\\)\ub294 MLM\uc758 \ube44\uc911\uc744 \uc870\uc815\ud558\ub294 \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130\uc785\ub2c8\ub2e4.<\/p>\n\n\n\n
    $$\\mathbb{L}_{PT} = \\mathbb{L}_{IIC} + \\lambda \\cdot \\mathbb{L}_{MLM}$$<\/p>\n\n\n\n
    Two-stage Finetuning <\/h3>\n\n\n\n
    Pre-training \ub2e8\uacc4\uc640 \ub9c8\ucc2c\uac00\uc9c0\ub85c two-stage finetuning \uacfc\uc815\uc5d0\uc11c\ub3c4 Recformer\uac00 direct \ud558\uac8c item\uc744 \uc778\ucf54\ub529\ud569\ub2c8\ub2e4. \uadf8\ub7ec\ub098, pre-training\uc5d0\uc11c\uc640\ub294 \ub2e4\ub974\uac8c in-batch negative \ubc29\uc2dd\uc774 \uc5ec\uae30\uc11c\ub294 \ubb38\uc81c\uac00 \ub420 \uc218\ub3c4 \uc788\uc2b5\ub2c8\ub2e4 (\ud2b9\ud788 \uc791\uc740 dataset\uc5d0\uc11c). \uc774\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 \uc800\uc790\ub294 item feature matrix \\(I \\in \\mathbb{R}^{|I|\\times{d}}\\) \ub97c \ub9cc\ub4dc\ub294\ub370, item embedding table\uacfc \ub2e4\ub978 \uc810\uc740 \\(I\\)\uac00 learnable\ud558\uc9c0 \uc54a\ub2e4\ub294 \uc810\uc785\ub2c8\ub2e4. Stage 1\uc5d0\uc11c \\(I\\)\ub97c epoch\ub9c8\ub2e4 \uc5c5\ub370\uc774\ud2b8 \ud558\uace0, stage 2\uc5d0\uc11c\ub294 freeze \ub41c \\(i\\)\ub97c \uac00\uc9c0\uace0 \ubaa8\ub378\uc758 parameter\ub9cc \ud559\uc2b5 \ubc0f \uc5c5\ub370\uc774\ud2b8 \ud558\ub294 two-stage finetuning \ubc29\ubc95\uc744 \ubc18\ubcf5\ud558\uba70 \ubaa8\ub378\uc740 \ucd5c\uc801\uc758 performance\uc5d0 \uac00\uae4c\uc6cc\uc9c0\uac8c \ub429\ub2c8\ub2e4. Finetuning\uc5d0\uc11c\ub3c4 IIC \ubc29\uc2dd\uc744 \uc0ac\uc6a9\ud558\uae34 \ud558\uc9c0\ub9cc, in-batch negative\uac00 \uc544\ub2cc fully softmax\ub97c \uc0ac\uc6a9\ud569\ub2c8\ub2e4. Two-stage Finetuning \uc54c\uace0\ub9ac\uc998\uc740 \uadf8\ub9bc 6\uc5d0\uc11c \ud655\uc778\ud558\uc2e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n
    \n
    \"\"
    \uadf8\ub9bc 6. Two-stage Finetuning Algorithm<\/figcaption><\/figure>\n<\/div>\n\n\n
    Experiment<\/h2>\n\n\n\n
    \uc2e4\ud5d8 \uc138\ud305<\/h3>\n\n\n\n
      \n
    • Pre-training\uacfc finetuning\uc744 \uc704\ud55c \ub370\uc774\ud130\uc14b\uc744 \uad6c\ucd95\ud558\uae30 \uc704\ud574 Amazon review dataset\uc758 \ub2e4\uc591\ud55c \uce74\ud14c\uace0\ub9ac\ub4e4\uc744 \ud65c\uc6a9\ud558\uc600\ub2e4\uace0 \ud569\ub2c8\ub2e4. <\/li>\n\n\n\n
    • Baseline\uc73c\ub85c\ub294 ID-Only methods, ID-Text methods, Text-Only methods \uce74\ud14c\uace0\ub9ac\ub85c \ub098\ub204\uc5b4 \ucd1d 8\uac1c\uc758 \uae30\uc874 SOTA \ubaa8\ub378\ub4e4\uacfc \ube44\uad50\ud558\uc600\uc2b5\ub2c8\ub2e4.<\/li>\n\n\n\n
    • Evaluation\uc73c\ub85c\ub294 NDCG@N, Recall@N, \uadf8\ub9ac\uace0 MRR\uc744 \ud65c\uc6a9\ud558\uc600\uace0, data splitting strategy\ub85c leave-one-out strategy\ub97c \uc0ac\uc6a9\ud588\uc2b5\ub2c8\ub2e4. <\/li>\n<\/ul>\n\n\n\n
      Overall Performance<\/h3>\n\n\n
      \n
      \"\"
      \uadf8\ub9bc 7. \ubca0\uc774\uc2a4\ub77c\uc778 \uc131\ub2a5 \ube44\uad50<\/figcaption><\/figure>\n<\/div>\n\n\n
      \uadf8\ub9bc 7 (Table 2)\ub97c \ubcf4\uba74, \ub300\ubd80\ubd84\uc758 \ubca4\uce58\ub9c8\ud06c\uc5d0\uc11c RecFormer\uac00 \uac00\uc7a5 \uc88b\uc740 \uc131\ub2a5\uc744 \ub0c8\uc74c\uc744 \uc54c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130\uc14b\uc5d0 \ub530\ub77c BERT4Rec\uacfc S3-Rec \ubaa8\ub378\uc774 \ube44\uad50\uc801 \uc88b\uc740 \uc131\ub2a5\uc744 \ubcf4\uc778 \uc810\ub3c4 \uc8fc\ubaa9\ud560\ub9cc \ud574\ubcf4\uc785\ub2c8\ub2e4. <\/p>\n\n\n
      \n
      \"\"
      \uadf8\ub9bc 8. Ablation study \uacb0\uacfc<\/figcaption><\/figure>\n<\/div>\n\n\n
      \uadf8\ub9bc 8 (Table 4)\uc5d0\uc11c \uac01\uac01\uc758 component\uac00 \ucd5c\uc885 prediction performance\uc5d0 \uc5bc\ub9c8\ub098 \uae30\uc5ec\ud588\ub294\uc9c0 \uc2e4\ud5d8\ud558\ub294 Ablation study \uacb0\uacfc\ub97c \ubcf4\uc5ec\uc8fc\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n\n\n\n
      Discussion<\/h2>\n\n\n\n
        \n
      1. \uae30\uc874 item ID\uc640 trainable item embedding\uc744 \ud65c\uc6a9\ud558\ub294 \uae30\uc874 \ubc29\uc2dd\ub4e4\ubcf4\ub2e4 \ub098\uc740 \uc810<\/strong>: \uc800\uc790\ub294 data sparsity \ubb38\uc81c\uc640 cold-start item \ubb38\uc81c\ub97c \ud6a8\uacfc\uc801\uc73c\ub85c \ud574\uacb0\ud560 \uc218 \uc788\ub2e4\uace0 \ud569\ub2c8\ub2e4.<\/li>\n\n\n\n
      2. \uae30\uc874 context-aware \ubc29\ubc95\uacfc\uc758 \ucc28\uc774\uc810<\/strong>: UniSRec, S3Rec, ZESRec\uacfc \uac19\uc740 Item ID\uc5d0 \ub300\ud55c \uc758\uc874\uc131\uc744 \ub0ae\ucd94\uae30 \uc704\ud55c context-aware sequential recommenders\ub3c4 \uae30\uc874\uc5d0 \uc788\uae34 \ud588\uc5c8\uc2b5\ub2c8\ub2e4. \uc774 \ubc29\ubc95\ub4e4\uc740 \uc8fc\ub85c side information\uc744 prior knowledge\ub85c \uc0ac\uc6a9\ud558\uae30 \uc704\ud574 BERT\uc640 \uac19\uc740 feature extractor\uc744 \ud1b5\ud574 item feature vector\ub97c \ucd94\ucd9c\ud558\uace0, item representation\uc5d0 fuse \uc2dc\ud0a4\ub294 \ubc29\uc2dd\uc744 \uc0ac\uc6a9\ud588\uc5c8\uc2b5\ub2c8\ub2e4. \uadf8\ub4e4\uacfc \ube44\uad50\ud588\uc744 \ub54c RecFormer\uc758 novelty\ub294 language representation\uc5d0 \uc9d1\uc911\ud55c new paradigm\uc744 \uc81c\uc548\ud588\ub2e4\ub294 \uac83\uc785\ub2c8\ub2e4. \uc800\uc790\ub294 natural language\ub97c \uc798 \ud65c\uc6a9\ud558\uc5ec new domain adaptation\uacfc cold-start item understanding \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370\uc5d0 \uc774\uc810\uc744 \uc5bb\uc744 \uc218 \uc788\ub2e4\uace0 \uc774\uc57c\uae30\ud569\ub2c8\ub2e4. <\/li>\n\n\n\n
      3. Cold-start problem<\/strong>: RecFormer\uc740 training set\uc5d0 \uc804\ud600 \ub4f1\uc7a5\ud558\uc9c0 \uc54a\uc740 item\ub3c4 item text\ub97c \ubc14\ub85c encoding \ud560 \uc218 \uc788\uae30\uc5d0 cold-start problem \ubb38\uc81c\uc5d0\uc11c \uac70\uc758 \uc790\uc720\ub86d\uc2b5\ub2c8\ub2e4. \ucd94\ucc9c\uc2dc\uc2a4\ud15c\uc5d0\uc11c main challenge\uc778 data sparsity\uc640 cold-start problem\uc744 language representation\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\uc2dd\uc73c\ub85c \ud6a8\uacfc\uc801\uc73c\ub85c, \ub610 \ube44\uad50\uc801 \uac04\ub2e8\ud558\uac8c \ud480\uc5b4\ub0c8\ub2e4\ub294 \uc810\uc774 \ud765\ubbf8\ub86d\uc2b5\ub2c8\ub2e4. <\/li>\n\n\n\n
      4. Transfer Learning for Recommendation<\/strong>: Source\uc640 target domain \uac04\uc758 shared information\uc744 \ud1b5\ud574 \uc9c0\uc2dd \uc804\ub2ec (knowledge transfer)\uc744 \ud558\uace0, universal item representation\uc744 \ud559\uc2b5\ud558\uae30 \uc704\ud574 additional layer\uc744 \ucd94\uac00\ud574\uc57c \ud588\ub358 \uae30\uc874 \ubc29\ubc95\ub4e4\uc740 \uc81c\uc57d\uc774 \ub9ce\uc740 \ubc18\uba74, \ud574\ub2f9 \ub17c\ubb38\uc5d0\uc11c\ub294 language representation\uc744 \ud6a8\uacfc\uc801\uc73c\ub85c \ud558\ub294 \ubc29\ubc95\uc73c\ub85c generality of natural language\ub97c \uc801\uadf9 \ud65c\uc6a9\ud558\uc5ec transfer knowledge\ub97c \ud558\uc600\uc2b5\ub2c8\ub2e4.<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"
        Sequential Recommendation \ud0dc\uc2a4\ud06c\uc5d0 \uad00\uc2ec\uc744 \uac00\uc9c0\uace0 \uc788\ub358 \uc640\uc911, 2023\ub144\ub3c4 KDD\ub97c \uae30\uc6c3\uac70\ub9ac\ub2e4\uac00 \ubc1c\uacac\ud55c \ub17c\ubb38 “Text Is All You Need: Learning Language Representations for Sequential Recommendation” \ub9ac\ubdf0\uc785\ub2c8\ub2e4. \uac04\ub2e8\ud574 \ubcf4\uc774\ub294 idea\ub85c \uc0c8\ub85c\uc6b4 sequential recommendation paradigm\uc744 \uace0\uc548\ud55c \uc7ac\ubbf8\uc788\ub294 \ub17c\ubb38\uc774\uc5c8\uc2b5\ub2c8\ub2e4. Keyword: Language Representation, Sequential recommendation, RecFormer, Transfer Learning […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[14,64],"tags":[37,66,65],"class_list":["post-368","post","type-post","status-publish","format-standard","hentry","category-natural-language-processing","category-recommender-system","tag-nlp","tag-paper-review","tag-recsys"],"yoast_head":"\n\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/jaeinlab.com\/?p=368\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab\" \/>\n<meta property=\"og:description\" content=\"Sequential Recommendation \ud0dc\uc2a4\ud06c\uc5d0 \uad00\uc2ec\uc744 \uac00\uc9c0\uace0 \uc788\ub358 \uc640\uc911, 2023\ub144\ub3c4 KDD\ub97c \uae30\uc6c3\uac70\ub9ac\ub2e4\uac00 \ubc1c\uacac\ud55c \ub17c\ubb38 “Text Is All You Need: Learning Language Representations for Sequential Recommendation” \ub9ac\ubdf0\uc785\ub2c8\ub2e4. \uac04\ub2e8\ud574 \ubcf4\uc774\ub294 idea\ub85c \uc0c8\ub85c\uc6b4 sequential recommendation paradigm\uc744 \uace0\uc548\ud55c \uc7ac\ubbf8\uc788\ub294 \ub17c\ubb38\uc774\uc5c8\uc2b5\ub2c8\ub2e4. Keyword: Language Representation, Sequential recommendation, RecFormer, Transfer Learning […]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/jaeinlab.com\/?p=368\" \/>\n<meta property=\"og:site_name\" content=\"Jaein's Lab\" \/>\n<meta property=\"article:published_time\" content=\"2024-01-10T09:01:26+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-01-11T09:43:52+00:00\" \/>\n<meta name=\"author\" content=\"Jaein\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Jaein\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368\"},\"author\":{\"name\":\"Jaein\",\"@id\":\"http:\\\/\\\/jaeinlab.com\\\/#\\\/schema\\\/person\\\/70edb94493e55ab8edbafc48bdf56883\"},\"headline\":\"\\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0)\",\"datePublished\":\"2024-01-10T09:01:26+00:00\",\"dateModified\":\"2024-01-11T09:43:52+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368\"},\"wordCount\":529,\"commentCount\":0,\"keywords\":[\"NLP\",\"paper review\",\"RecSys\"],\"articleSection\":[\"Natural Language Processing\",\"Recommender System\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/jaeinlab.com\\\/?p=368#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368\",\"url\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368\",\"name\":\"\\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab\",\"isPartOf\":{\"@id\":\"http:\\\/\\\/jaeinlab.com\\\/#website\"},\"datePublished\":\"2024-01-10T09:01:26+00:00\",\"dateModified\":\"2024-01-11T09:43:52+00:00\",\"author\":{\"@id\":\"http:\\\/\\\/jaeinlab.com\\\/#\\\/schema\\\/person\\\/70edb94493e55ab8edbafc48bdf56883\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/jaeinlab.com\\\/?p=368\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/jaeinlab.com\\\/?p=368#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"http:\\\/\\\/jaeinlab.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0)\"}]},{\"@type\":\"WebSite\",\"@id\":\"http:\\\/\\\/jaeinlab.com\\\/#website\",\"url\":\"http:\\\/\\\/jaeinlab.com\\\/\",\"name\":\"Jaein's Lab\",\"description\":\"Natural Language Processing, Knowledge Graph, and others!\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"http:\\\/\\\/jaeinlab.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Person\",\"@id\":\"http:\\\/\\\/jaeinlab.com\\\/#\\\/schema\\\/person\\\/70edb94493e55ab8edbafc48bdf56883\",\"name\":\"Jaein\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g\",\"caption\":\"Jaein\"},\"description\":\"I'm a Computer Science PhD candidate from South Korea, currently based in Beijing, China. My academic journey is dedicated to pioneering research in Natural Language Processing and Knowledge Graphs, striving to break new ground in these dynamic fields.\",\"sameAs\":[\"http:\\\/\\\/jaeinlab.com\",\"https:\\\/\\\/www.linkedin.com\\\/in\\\/jaeinlab\\\/\"],\"url\":\"https:\\\/\\\/jaeinlab.com\\\/?author=1\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/jaeinlab.com\/?p=368","og_locale":"en_US","og_type":"article","og_title":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab","og_description":"Sequential Recommendation \ud0dc\uc2a4\ud06c\uc5d0 \uad00\uc2ec\uc744 \uac00\uc9c0\uace0 \uc788\ub358 \uc640\uc911, 2023\ub144\ub3c4 KDD\ub97c \uae30\uc6c3\uac70\ub9ac\ub2e4\uac00 \ubc1c\uacac\ud55c \ub17c\ubb38 “Text Is All You Need: Learning Language Representations for Sequential Recommendation” \ub9ac\ubdf0\uc785\ub2c8\ub2e4. \uac04\ub2e8\ud574 \ubcf4\uc774\ub294 idea\ub85c \uc0c8\ub85c\uc6b4 sequential recommendation paradigm\uc744 \uace0\uc548\ud55c \uc7ac\ubbf8\uc788\ub294 \ub17c\ubb38\uc774\uc5c8\uc2b5\ub2c8\ub2e4. Keyword: Language Representation, Sequential recommendation, RecFormer, Transfer Learning […]","og_url":"https:\/\/jaeinlab.com\/?p=368","og_site_name":"Jaein's Lab","article_published_time":"2024-01-10T09:01:26+00:00","article_modified_time":"2024-01-11T09:43:52+00:00","author":"Jaein","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Jaein","Est. reading time":"6 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/jaeinlab.com\/?p=368#article","isPartOf":{"@id":"https:\/\/jaeinlab.com\/?p=368"},"author":{"name":"Jaein","@id":"http:\/\/jaeinlab.com\/#\/schema\/person\/70edb94493e55ab8edbafc48bdf56883"},"headline":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0)","datePublished":"2024-01-10T09:01:26+00:00","dateModified":"2024-01-11T09:43:52+00:00","mainEntityOfPage":{"@id":"https:\/\/jaeinlab.com\/?p=368"},"wordCount":529,"commentCount":0,"keywords":["NLP","paper review","RecSys"],"articleSection":["Natural Language Processing","Recommender System"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/jaeinlab.com\/?p=368#respond"]}]},{"@type":"WebPage","@id":"https:\/\/jaeinlab.com\/?p=368","url":"https:\/\/jaeinlab.com\/?p=368","name":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0) - Jaein's Lab","isPartOf":{"@id":"http:\/\/jaeinlab.com\/#website"},"datePublished":"2024-01-10T09:01:26+00:00","dateModified":"2024-01-11T09:43:52+00:00","author":{"@id":"http:\/\/jaeinlab.com\/#\/schema\/person\/70edb94493e55ab8edbafc48bdf56883"},"breadcrumb":{"@id":"https:\/\/jaeinlab.com\/?p=368#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/jaeinlab.com\/?p=368"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/jaeinlab.com\/?p=368#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"http:\/\/jaeinlab.com\/"},{"@type":"ListItem","position":2,"name":"\bText Is All You Need? \uc5b8\uc5b4 \ud45c\ud604\uc73c\ub85c \uc21c\ucc28 \ucd94\ucc9c\uc2dc\uc2a4\ud15c \ub9cc\ub4e4\uae30 (\ud398\uc774\ud37c \ub9ac\ubdf0)"}]},{"@type":"WebSite","@id":"http:\/\/jaeinlab.com\/#website","url":"http:\/\/jaeinlab.com\/","name":"Jaein's Lab","description":"Natural Language Processing, Knowledge Graph, and others!","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"http:\/\/jaeinlab.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Person","@id":"http:\/\/jaeinlab.com\/#\/schema\/person\/70edb94493e55ab8edbafc48bdf56883","name":"Jaein","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/6a23798f753744134b0f342a342b0fb0d4035a7313d8b7c54faf930978eaa942?s=96&d=wp_user_avatar&r=g","caption":"Jaein"},"description":"I'm a Computer Science PhD candidate from South Korea, currently based in Beijing, China. My academic journey is dedicated to pioneering research in Natural Language Processing and Knowledge Graphs, striving to break new ground in these dynamic fields.","sameAs":["http:\/\/jaeinlab.com","https:\/\/www.linkedin.com\/in\/jaeinlab\/"],"url":"https:\/\/jaeinlab.com\/?author=1"}]}},"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"","_links":{"self":[{"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/posts\/368","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=368"}],"version-history":[{"count":6,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/posts\/368\/revisions"}],"predecessor-version":[{"id":471,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=\/wp\/v2\/posts\/368\/revisions\/471"}],"wp:attachment":[{"href":"https:\/\/jaeinlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=368"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=368"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jaeinlab.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=368"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}