In this paper we address the problem of inferring social structure and dominance relationships in a group of rhesus macaques (a species of monkey) using only position data captured using RFID tags. Automatic inference of the social structure in an animal group enables a number of important capabilities, including: 1) A verifiable measure of how the social structure is affected by an intervention such as a change in the environment, or the introduction of another animal, and 2) A potentially significant reduction in person hours normally used for assessing these changes. Social structure in a group is an important indicator of its members relative level of access to resources and has interesting implications for an individuals health and learning in groups. There are two main quantitative criteria assessed in order to infer the social structure; Time spent close to conspecifics, and displacements. An interaction matrix is used to represent the total duration of events detected as grooming behavior between any two monkeys. This forms an undirected tie-strength (closeness of relationships) graph. A directed graph of hierarchy is constructed by using the well cited assumption of a linear hierarchy for rhesus macaques. Events that contribute to the adjacency matrix for this graph are withdrawals or displacements where a lower ranked monkey moves away from a higher ranked monkey. Displacements are one of the observable behaviors that can act as a strong indication of tie-strength and dominance. To quantify the directedness of interaction during these events we construct histograms of the dot products of motion orientation and relative position. This gives us a measure of how much time a monkey spends in moving towards or away from other group members.
Download