SESSIONINTENTBENCH
收藏资源简介:
SESSIONINTENTBENCH是一个多任务跨会话意图转换建模基准,用于电子商务客户行为理解。该数据集包含1952177个意图条目,1132145个会话意图轨迹,以及13003664个可用任务。数据集来源于对10,905个会话的挖掘,并经过人类注释以收集部分数据的真实标签,形成一个评估金集。该数据集旨在评估语言模型在理解会话意图转换方面的能力,并为电子商务产品购买会话中的意图建模提供可扩展的方法。
SESSIONINTENTBENCH is a multi-task cross-session intent transition modeling benchmark dedicated to e-commerce customer behavior understanding. This dataset contains 1,952,177 intent entries, 1,132,145 session intent trajectories, and 13,003,664 available tasks. It is sourced from the mining of 10,905 sessions, and partial data has been manually annotated to gather ground-truth labels, thereby creating a gold standard evaluation set. This benchmark is designed to assess the capability of language models in comprehending session intent transitions, and offer scalable approaches for intent modeling during e-commerce product purchase sessions.

- 1SessionIntentBench: A Multi-task Inter-session Intention-shift Modeling Benchmark for E-commerce Customer Behavior Understanding香港科技大学计算机科学与工程系, 亚马逊公司 · 2025年



