Abstract | ||
---|---|---|
We propose and evaluate three novel approaches to the problem of performing similarity matching on full-length event-interval sequences (e-sequences). The ERF approach represents each e-sequence as a vector of the magnitudes of the durations of the events. Euclidean distances between the vectors are then used to compare given e-sequences. The EPC approach embeds each e-sequence as a vector of position codes to capture the order of the occurrences of events explicitly and temporal relations among the events implicitly. Cosine distances between the vectors are used to infer similarity of e-sequences. Finally, the EMKL approach combines the ERF and EPC approaches using multiple kernel learning. Empirical evaluation on eight real datasets suggests that the EMKL approach outperforms existing state-of-the-art methods in terms of nearest neighbor classification accuracy. |
Year | DOI | Venue |
---|---|---|
2020 | 10.1007/978-3-030-47358-7_43 | Canadian Conference on AI |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
S. Mohammad Mirbagheri | 1 | 0 | 0.34 |
Howard J. Hamilton | 2 | 1501 | 145.55 |