GroupSense: A Lightweight Framework for Group Identification

Publications by: Snigdha Das, Soumyajit Chatterjee, Sandip Chakraborty, Bivas Mitra
GroupSense: A Lightweight Framework for Group Identification TMC'19 [Full Paper] [IEEE]
An Unsupervised Model for Detecting Passively Encountering Groups from WiFi Signals GLOBECOM'18 [Full Paper] [IEEE] [Slides]

Meeting Group

In an organization, individuals prefer to form various formal and informal groups for mutual interactions. Therefore, ubiquitous identification of such groups and understanding their dynamics are important to monitor activities, behaviours and well-being of the individuals. Meeting groups are the groups where co-located group members occasionally interact with each other. Precisely, given a population of subjects U, we define a meeting group G[t,t+T] ⊆ U for the time period [t, t+T ] as the collection of co-located individuals {u ∈ U} sharing similar context.



  • Colocation

    Identification of a location of the meeting group can be conceptualized as a localization problem. However, retrieving highly precise location information is challenging.

  • Context

    User list are not predefined in physical world group. Therefore, getting prior context of the users is challenging.



In this project, we develop a lightweight, yet near-accurate, methodology, called GroupSense, to identify various interacting groups based on collective sensing through users’ smartphones.


Proposed Solution: GroupSense

Primary Indicators:

  • Overlapping WiFi access points exhibit similarity for subjects belonging to the same meeting group.
  • The subjects participating in the same group exhibit similar audio signature.


  • WiFi signal strength is highly similar for subjects with same group as compared to the different group.
  • Due to device heterogenity, a wide variation in signal strength across smartphone models and devices present.
  • In noisy environment, the frequency component of the audio signal deviates even for the same group subjects.
  • Device heterogenity exhibits more audio amplitude similarity for different group subjects than the same group subjects.

Admitted Steps in GroupSense :

  • Overlapping WiFi access points are measured using Jaccard Coefficient to detect the proximity.
  • Gain factor computed using WiFi signal strength is used to eliminate the device heterogenity effect.
  • Acknowledging the effect of noisy environment and device herterogenity, the audio tone of the subjects in the group are extracted by calculating Complex Cepstrum.
  • Weighted proximity-audio features are computed and fed to the WalkTrap community detection algorithm for detect the meeting groups.

GroupSense Performance:

  • The proposed method has been implemented and tested under many real-life scenarios in an academic institute environment.
  • GroupSense can identify user groups with on an average 0.9(±0.14) F1-Score even in a noisy environment.