Abstract | ||
---|---|---|
Knowing individual real attributes is beneficial to intelligent marking, automatic advertising as well as a new application of smart home system. In this paper, we leverage seemingly innocent user information (step count) coming from ubiquitous mobile and sensing device for attribute inference. The utilization of sensor data breaks the traditional dependence on text, social relationship and online behavior in social media, and avoids the problem of usersu0027 gender variation in linguistic style on different social networks. Existing methods on attribute inference usually ignore the temporal information of data and purely explore different types of features manually. However, loss of temporal characteristic may reduce the representation ability to infer useru0027s attributes. Meanwhile, excessively depending on manually defined features requires a great mass of human labor and often suffers from under specification. To address these problems, we propose a novel Hybrid Multiple Representations (HMR) model that combines simple human knowledge with automated deep learning by using stacked bidirectional LSTM, Bag-of-step and Holiday Activation methods to predict gender and age from users daily step count. Experiments are conducted on a real-world pedometer dataset where gender and age are to be predicted. The empirical results show that our HMR model does a good job on the task of predicting attributes compared with state-of-the-art baselines. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/SmartWorld-UIC-ATC-SCALCOM-IOP-SCI.2019.00173 | SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 6 |