A co-learning framework for learning user search intents from rule-generated training data
Proceedings of the 33rd international ACM SIGIR conference on Research and development in information retrieval
| Hi-index | 0.00 |
By examining searcher behavior on a large search engine, we have identified seven basic kinds of task behaviors that can be observed in web search session logs. In the studies reported, we first manually labeled 700 complete web sessions, and then subsequently had 23 searchers self-label 252 days of their own sessions to give an accurate picture of what kinds of tasks people are doing when they search. From these two studies, we have found that the most accurate labeling of search task session data is done by the searchers themselves, and that it is very difficult for an external observer or automatic classifier to infer where the task boundaries are or what the actual user task goal is.