@inproceedings {pub4335,
	title = {Target-aware Prediction of Tool Usage in Sequential Repair Tasks},
	author = {Nima Nabizadeh AND Martin Ernst Heckmann AND Dorothea Kolossa},
	year = {2020},
	abstract = {Many real-world tasks have a sequential nature comprising several steps, and in a complex repair task, each step might involve a different tool. Learning the sequential pattern of tool usage would be helpful for various assistance scenarios, e.g.~allowing a contextualized assistant to predict the next required tool in an unseen task. In this work, we examine the potential of this idea on an example of sequential tasks: repairing electronic devices. We employ two prominent classes of sequence learning methods for modeling the tool usage, including Variable Order Markov Models (VMMs) and Recurrent Neural Networks (RNNs). We then extend these methods to be conditioned on extra information that exists in repair manuals and represents the repair target, including the name of the component that needs to be repaired, and the category of the device. We investigate the effect of target-awareness and long-term dependencies on the prediction performance and compare the methods in terms of accuracy and training time. The evaluation using an annotated dataset of repair manuals shows that both target-awareness and long-term dependencies have a substantial effect on the tool prediction. While the RNN has slightly more accurate predictions in most scenarios, the VMM has a lower training time and is beneficial when the prediction needs to be restricted with respect to the device category.
},
	publisher = {Springer, LNCS, LNAI, LNBI},
	booktitle = {Machine Learning, Optimization, and Data Science},
	editor = {Giuseppe Nicosia,Varun Ojha,Emanuele La Malfa,Giorgio Jansen,Vincenzo Sciacca,Panos M. Pardalos,Giovanni Giuffrida,Renato Umeton},
	city = {Siena},
	edition = {1},
	volume = {2},
	pages = {869{\textendash}880},
	series = {Information Systems and Applications, including Internet / Web, and HCI}
}