Reality mining of animal social systems

Jens Krause, Stefan Krause, Robert Arlinghaus, Ioannis Psorakis, Stephen Roberts, Christian RutzTrends 2013 · DOI

TL;DR

This paper introduces "reality mining" – the application of machine-sensed data to understand animal social behavior – as a powerful new approach for studying animal societies. By leveraging advancements in miniaturized tracking technology, researchers can now gather unprecedentedly detailed data on entire populations, shifting the focus from data collection to sophisticated analysis and modeling of complex social dynamics.

What Problem Does This Paper Address?

Studying the intricate social lives of animals in their natural habitats has always been a significant challenge. Traditional methods, like direct observation or infrequent resighting, provide only snapshots of behavior and are often limited to small sample sizes or specific individuals. This makes it difficult to understand the dynamic processes that shape social structures, such as how social hierarchies emerge, how information flows, or how diseases spread within a population. The paper addresses this gap by proposing "reality mining," an approach that uses automated, machine-sensed data to capture near-continuous social interactions at a population level, offering a much richer and more detailed understanding of animal social systems.

Methods

The paper outlines two primary conceptual approaches for collecting social interaction data:

  • Direct Encounter Mapping: This method uses animal-borne proximity loggers (often called "business card tags" or "transceiver tags") that communicate directly with each other. When tagged individuals come within a certain range, their tags record the encounter. This allows for high-resolution, reciprocal recording of social interactions. Some systems can also record Received Signal Strength Indicator (RSSI) values, which can be calibrated to estimate animal-to-animal distances.
  • Indirect Encounter Mapping: This approach relies on charting the spatiotemporal positions and movements of individuals. This can be done using technologies like GPS loggers or acoustic telemetry, where individual locations are recorded. Encounters are then inferred by analyzing co-occurrence data at specific locations or within defined time windows. Passive Integrated Transponder (PIT) tags and Radio-Frequency Identification (RFID) systems are also used, where fixed readers detect the presence of tagged animals at specific sites.

For data analysis, the paper highlights two key techniques:

  • Social Network Analysis (SNA): This method visualizes and analyzes relationships between individuals (nodes) based on their associations (edges). SNA allows for the measurement of individual characteristics and global network properties, providing insights into social structure and information flow.
  • Hidden Markov Models (HMMs): HMMs are powerful statistical tools for modeling systems with unobservable states. In this context, they can infer behavioral states (e.g., "social," "foraging") from observable data (e.g., location, proximity) and model the transitions between these states. This is particularly useful for understanding the dynamics of social behavior and identifying routines.

Datasets

The paper does not introduce specific new datasets for training or evaluation. Instead, it discusses how existing and newly generated data from animal-borne tracking devices can be utilized. The examples provided draw on data from studies on New Caledonian crows, great tits, and other wild bird populations, as well as research on sharks, cattle, and possums.

Key Insights and Ideas

  • Bridging the Gap: The miniaturization of tracking technology is enabling biologists to adopt "reality mining" approaches, mirroring human studies that use digital footprints. This allows for unprecedented population-level data collection.
  • Dynamic Social Structures: Reality mining can reveal how social structures emerge and change dynamically over different timescales, moving beyond static snapshots of behavior.
  • Beyond Proximity: While current methods primarily detect physical proximity, future advancements with integrated sensors (video, accelerometers, physiological sensors) will provide crucial contextual information about the nature of interactions.
  • The Power of "Zero" Edges: Proximity logging, in particular, can confidently identify the absence of social encounters ("zero edges"), which is vital for robust social network analysis and understanding avoidance behaviors.
  • HMMs for Behavioral Dynamics: HMMs offer a sophisticated way to model the underlying behavioral dynamics that generate observed social patterns, allowing for comparisons across groups and environments.

Key Results

The paper doesn't present specific quantitative results from a single study but rather synthesizes findings from various "reality mining" applications. Key qualitative results and potentials highlighted include:

  • Detailed Social Networks: Studies have successfully mapped social networks in wild bird populations, revealing substantial close-range associations and potential for rapid information flow (New Caledonian crows).
  • Inferring Behavior from Movement: Indirect methods, like PIT tags and RFID readers, have been used to study social dynamics in great tits, enabling the study of pair bond formation and foraging behavior.
  • Population-Level Insights: Reality mining allows for the study of entire local populations, providing a more representative understanding of social dynamics than traditional methods.
  • Understanding Disease Transmission: By mapping contact rates and social structures, reality mining can provide crucial data for modeling disease spread in animal populations.

Limitations and Future Work

  • Tag Limitations: Current proximity loggers, while miniaturized, can still be relatively large for smaller species and have limited battery and memory lifespans.
  • Geospatial Referencing: Basic direct encounter mapping often lacks geospatial information, requiring integration with other technologies like GPS for detailed movement analysis.
  • Contextual Information: A major limitation is the lack of contextual information for encounters; current systems primarily detect proximity, not the nature or biological context of the interaction.
  • Data Management: The sheer volume of data generated by reality mining presents significant challenges for processing, screening, and analysis.
  • Calibration: Accurate conversion of raw sensor data (e.g., RSSI) into meaningful biological metrics requires laborious field calibration.

Future work will focus on:

  • Sensor Integration: Incorporating additional sensors into tracking devices to capture behavioral context, physiological states, and environmental factors.
  • Advanced Analytical Tools: Further development and application of sophisticated analytical techniques like HMMs and advanced social network analysis to extract deeper insights from complex datasets.
  • Cost-Effectiveness: Reducing the cost of technology and data processing to make reality mining more accessible for a wider range of research questions and species.
  • Ecological Validity: Combining reality mining data with experimental manipulations to establish causal links between social dynamics and ecological processes.

Summary updated

BibTeX
@inproceedings{krause2013reality,
  author = {Jens Krause and Stefan Krause and Robert Arlinghaus and Ioannis Psorakis and Stephen Roberts and Christian Rutz},
  title = {{Reality mining of animal social systems}},
  booktitle = {Trends},
  year = {2013},
  doi = {10.1016/j.tree.2013.06.002},
}
Figure