Evaluation of Session Segmentation Methods Using Behavior and Text Embeddings
246 Views
November 02, 24
スライド概要
Yongzhi Jin, Kazushi Okamoto: Evaluation of session segmentation methods using behavior and text embeddings, The 11th International Symposium on Computational Intelligence and Industrial Applications (ISCIIA2024), 2024.11, Beijing, PRC.
Data Science Research Group, The University of Electro-Communications
関連スライド
各ページのテキスト
:Session :Item 1 1 2 3 2 6 7 8 3 9 10 ? 4 5
1 2
Word Word2Vec Item2Vec
Segment Method Embedding Cos Similarity Item2Vec Session Set Word2Vec text-embedding-ada-002 k-means PR-AUC ROC-AUC F1-score LightGBM Evaluation SVM LR Annotation
Item 1 Item 2 Item 3 Item 4 Item 5 Embedding 3 Item 3 'Item Name' Pre-trained model Text Embedding Embedding 3
Segmentation Input YES Segmentation Cosine Similarity LightGBM SVM LR Input k-means YES Embed Embed Embed Embed Cluster A Cluster A Cluster B Cluster B Item 1 Item 2 Item 3 Item 4 Item 1 Item 2 Item 3 Item 4 Output : Segment probability {0.00 1.00} Output : Segment judgment {0,1}