解釈・制御可能なMulti-Interest推薦モデルの提案

>100 Views

June 08, 26

スライド概要

小西 渓士郎, 岡本 一志, 軽部 幸起, 原田 慧, 柴田 淳司: 解釈・制御可能なMulti-Interest推薦モデルの提案, 第40回人工知能学会全国大会, 2026.6, 群馬県高崎市.

profile-image

Data Science Research Group, The University of Electro-Communications

シェア

またはPlayer版

埋め込む »CMSなどでJSが使えない場合

ダウンロード

関連スライド

各ページのテキスト
1.

Multi-Interest 2026.06.08 2026 1 / 21

2.

2026.06.08 2026 2 / 21

3.

1. 2. Multi-Interest 2026.06.08 2026 3 / 21

4.

Multi-Interest 2026.06.08 2026 4 / 21

5.

Multi-Interest MIND [Li+, 2019] Dynamic Routing × × ComiRec-SA [Cen+, 2020] Attention × × TimiRec [Wang+, 2022] Attention × × Attention 2026.06.08 2026 5 / 21

6.

→ RQ.1: RQ.2: 2026.06.08 2026 6 / 21

7.

[1/4] 2026.06.08 2026 7 / 21

8.

[2/4] MLP 2026.06.08 2026 8 / 21

9.

Task1 MLP [3/4] α α Sampled Softmax Cross-Entropy Loss 2026.06.08 2026 9 / 21

10.

Task2 [4/4] Load Balancing Bias Binary Cross Entropy 2026.06.08 2026 10 / 21

11.

2026.06.08 2026 11 / 21

12.

MovieLens 1M EC Yelp MovieLens 1M Yelp 6,040 69,912 3,706 14,480 1,000,209 713,662 19 2026.06.08 2026 1026 12 / 21

13.

Hit@ NDCG@ TargetRate@ 2026.06.08 2026 13 / 21

14.

GRU4Rec [Hidasi+, 2015] Single Interest RNN SASRec[Kang+, 2018] Single Interest Transformer BERT4Rec[Sun+, 2019] Single Interest MIND [Li+, 2019] Multi-Interest ComiRec-SA[Cen+, 2020] Multi-Interest 2026.06.08 2026 14 / 21

15.

MovieLens 1M Metric RQ.1 [1/2] GRU4Rec BERT4Rec SASRec MIND ComiRec-SA HIT@1 0.220 0.181 0.217 0.141 0.169 0.178 HIT@5 0.464 0.434 0.459 0.400 0.388 0.465 HIT@10 0.607 0.563 0.580 0.558 0.562 0.601 NDCG@5 0.346 0.315 0.344 0.272 0.281 0.327 NDCG@10 0.392 0.357 0.383 0.322 0.337 0.371 2026.06.08 2026 15 / 21

16.

Yelp RQ.1 Metric [2/2] GRU4Rec BERT4Rec SASRec MIND ComiRec-SA HIT@1 0.210 0.154 0.256 0.164 0.200 0.227 HIT@5 0.487 0.349 0.492 0.391 0.420 0.479 HIT@10 0.595 0.518 0.646 0.530 0.544 0.613 NDCG@5 0.354 0.249 0.381 0.280 0.313 0.358 NDCG@10 0.390 0.303 0.433 0.325 0.352 0.402 Yelp MovieLens 1M RQ.1 2026.06.08 2026 16 / 21

17.

RQ.2 α 95% MovieLens 1M 2026.06.08 1.47 2026 Yelp 1.73 RQ.2 17 / 21

18.

2026.06.08 2026 18 / 21

19.

Multi-Interest RQ.1 MovieLens 2026.06.08 1.47 2026 Yelp 1.73 RQ.2 19 / 21

20.

[Cen+, 2020]Y. Cen, J. Zhang, X. Zou, C. Zhou, H. Yang, J. Tang, ``Controllable Multi-Interest Framework for Recommendation", in Proc. 26th ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., 2942–2951, 2020 [Li+, 2019] C. Li, Z. Liu, M. Wu, Y. Xi, H. Zhao, P. Huang, G. Kang, Q. Chen, W. Li, D. Lun Lee, ``Multi-interest network with dynamic routing for recommendation at Tmall,'' in Proc. 28th ACM Int. Conf. Inf. Knowl. Manag., 2615--2623, 2019. [Wang+, 2022] C. Wang, Z. Wang, Y. Liu, Y. Ge, W. Ma, M. Zhang, Y. Liu, J. Feng, C. Deng, and S. MaA, ``Target Interest Distillation for Multi-Interest Recommendation,'' in Proc. 31st ACM Int. Conf. Inf. Knowl. Manag., 2007-2016, 2022. [Li+, 2022] J. Li, J. Zhu, Q. Bi, G. Cai, L. Shang, Z. Dong, X. Jiang, and Q. Liu, ``MINER: Multi-Interest Matching Network for News Recommendation,'' in Find. Assoc. for Comput. Lnguist.: ACL 2022, 343--352, 2022. [Kouki+, 2019] P. Kouki, J. Schaffer, J. Pujara, J. O'Donovan, L.Getoor, ``Personalized explanations for hybrid recommender systems,'' in Proc. 24th Int. Conf. Intell. User Interfaces, 379--390, 2019. [Hidasi+, 2015] B. Hidasi, A. Karatzoglou, L. Baltrunas, D. Tikk, ``Session-based recommendations with recurrent neural networks,'' arXiv preprint, arXiv:1511.06939, 2015. 2026.06.08 2026 20 / 21

21.

[Sun+, 2019] F. Sun, J. Liu, J. Wu, C. Pei, X. Lin, W. Ou, P. Jiang ``BERT4Rec: Sequential recommendation with bidirectional encoder representations from Transformer,'' in Proc. 28th ACM Int. Conf. Inf. Know. Manag., 1441-1450, 2019. [Kang+, 2018] W. Kang, J. McAuley, ``Self-attentive sequential recommendation,'' arXiv preprint, arXiv:1808.09781, 2018. 2026.06.08 2026 21 / 21