Abstract

User-oriented machine learning strategies for information extraction: Putting the human back in the loop. David Pierce and Claire Cardie. In Working Notes of the IJCAI-2001 Workshop on Adaptive Text Extraction and Mining, pages 80-81, 2001.

This position paper outlines a user-oriented meta-learning strategy that attempts to balance accuracy, coverage, and responsiveness when training examples are solicited from a real user. The key idea behind user-oriented learning is the recognition of the complementary strengths of the human user and the machine learner. On one hand, the human is proficient at judging an information structure as desirable or undesirable; on the other, the machine is proficient at rapidly locating similar examples from large quantities of text. Both human and machine are allowed to exercise their strengths in a scenario where the human begins by providing training examples, the machine attempts to locate additional instances based on the training examples, the human responds by confirming the desirability (or undesirability) of the new instances, the machine adds them to the training examples and continues to search for more instances, and so on.