The Paradox of Personalization: Does Task Prediction Require Individualized Models?
The Paradox of Personalization: Does Task Prediction Require Individualized Models?
复制标题
个性化的悖论：任务预测是否需要个性化模型？
DOI：
10.1145/3176349.3176887
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
C. Shah
中科院分区：
文献类型：
--
作者：
M. Mitsui;Jiqun Liu;C. Shah
我们探讨了以下两个方面之间的差距：1）任务和浏览行为之间的统计学显著关系; 2）从这些行为预测任务类型。先前的文献已经显示了Web浏览行为和人的相应搜索任务之间的关系。我们发现统计上显着的浏览器功能检测任务-比较功能，以前的文献-并将此知识应用于搜索会话的任务分类。尽管显著的特征改善了基线的预测，但并没有太大的改善。我们认为，更微妙的治疗这些功能应该超越统计学意义。在某些情况下，可能需要考虑个人模式才能进行有效的预测。
We explore the gap between 1) statistically significant relationships between task and browsing behavior and 2) predicting task type from such behaviors. Previous literature has shown relationships between Web browsing behavior and person»s corresponding search task. We find statistically significant browser features for detecting task - comparing the features to previous literature - and apply this knowledge to task classification of search sessions. Even though significant features improve prediction over baselines, it is not by much. We suggest that a more subtle treatment of such features should go beyond statistical significance. In some cases, considering personal patterns may be required for effective prediction.