In most social network studies, it is assumed that nodes are simple and carry no information, and links are explicit ties such as friendship. Which nodes are in which group is determined as a function of these explicit ties. For example, given a set of random walks through the network, it is possible to learn a vector for each node which contains a latent representation of the node. These latent representations have useful properties that can be easily exploited by statistical models for tasks like identifying groups and inferring implicit links. However, most existing representation learning methods ignore node attributes. In many cases, there is a rich body of information and events associated with nodes that also can be used for node clustering and to infer ties. In social media, e.g., an explicit relationship is friendship, and another is the follower-followee relation. Besides, there are the set of messages passed by the users, as well as, their activities in the form of liking or mentioning. What is needed is a way of collectively using both the explicit ties and this rich body of additional information in learning these latent node representations. Combining such data should enable more effective link inference and grouping strategies. In this research, we propose People2Vec an algorithm to learn representations that takes into account proximity between users due to their social media activities. We validate our model by experiments on two different social-media datasets and find the model to perform better than prior state-of-the-art approaches.