Determining user needs through abnormality detection and heterogeneous embedding of usage sequence

In this study, we propose an advanced method for determining user needs based on abnormality detection and heterogeneous embedding of the usage sequences. We focus on the implied needs at the fine-grained levels based on the usage sequence, whereas previous textual review-based approaches have focused on the explicit needs at the product levels. Moreover, although previous studies regarding a usage sequence have primarily focused on an analysis of the tendency, app prediction, or recommendations, we first attempted to uncover abnormal sequences regarding user needs. Furthermore, in terms of the methodology, we then attempted a heterogeneous embedding approach to calculate the vector representation of each element of the usage sequence including the application, buttons, content, or system keys by utilizing the metapath2vec algorithm, which differs from previous studies that have focused solely on the embedding application usage. Further, to apply the abnormality detection method in determining an abnormal sequence corresponding to the user needs, we calculate the vector representation of the entire usage sequence utilizing RNN-AE based on heterogeneous embedding. After examining and evaluating the extracted abnormal sequences with the help of domain experts from LG Electronics, the experimental results verify that our proposed method can effectively extract a meaningful abnormal sequence corresponding to the implied needs. In addition, we calculated the correlation of the coefficient between the abnormality score and the importance score of the extracted sequences to compare the performance of each sequence model and the abnormality detection method.