The WEKA data mining software: an update
ACM SIGKDD Explorations Newsletter
LIBSVM: A library for support vector machines
ACM Transactions on Intelligent Systems and Technology (TIST)
Hi-index | 0.00 |
Recent work on consultations between out-patients with schizophrenia and psychiatrists has shown that adherence to treatment can be predicted by patterns of repair -- specifically, the pro-activity of the patient in checking their understanding, i.e. patient clarification. Using machine learning techniques, we investigate whether this tendency can be predicted from high-level dialogue features, such as backchannels, overlap and each participant's proportion of talk. The results indicate that these features are not predictive of a patient's adherence to treatment or satisfaction with the communication, although they do have some association with symptoms. However, all these can be predicted if we allow features at the word level. These preliminary experiments indicate that patient adherence is predictable from dialogue transcripts, but further work is necessary to develop a meaningful, general and reliable feature set.