*4.1. Data Description*

We evaluated the solution using a public dataset [35], which is referred to as *StudentLife*. Wang et al. [35] performed during 66 days a passive sensing of social activities (i.e., conversations) derived from microphone data gathered from smartphones of 48 Dartmouth College undergraduate and graduate students. Conversation samples are composed of two fields: start and end timestamps of conversations experienced by participants. All collected data was anonymized to preserve privacy of the monitored individuals.

The used dataset contains conversation samples composed of two features: start and end timestamps of social interactions. Figure 8 shows the first lines of the dataset of an individual. For example, the second line in this file records that he/she experienced a conversation that started at Unix timestamp 1364359600 and ended at Unix timestamp 1364359812.



**Figure 8.** Dataset features used in the solution.

Firstly, we performed a data cleaning process to remove users who had insufficient data to conduct experiments. Only users who contained at least 52 days of collected data (≈80% of the study days) are in this experiment. We used data from 24 individuals who had sufficient data.

We represent each record as a social event to design a proper data flow for the proposed processing network. For this, we derived the following information from social activity records: social activity type (i.e., conversation), start time, and a set of CAs. The first step was to convert the Unix timestamp to a Date Java object to represent the event start time attribute and extract CAs. We identified the weekday when events occurred using their timestamps, so enabling to specify temporal context scales. We defined two context scales: a fine-grained scale composed of weekdays (e.g., Monday, Tuesday, and Wednesday) and a broad one to distinguish weekends (i.e., Saturdays and Sundays) from midweeks (i.e., Monday to Friday). In the end, the structure of the generated social event flows was as follows:

1 event = (activity: Conversation, start: 12:20, contexts: [Saturday, Weekend])

