Social Networks Agentbased Modelling and Social Network Analysis
Social Networks: Agent-based Modelling and Social Network Analysis with PAJEK Richard Taylor¹ and Gindo Tampubolon² ¹Centre for Policy Modelling, Manchester Metropolitan University ²Centre for Research on Innovation and Competition, University of Manchester ESRC Research Methods Festival, Oxford, 17 th-20 th July 2006, & Oxford Spring School, Dept. of Politics and International Relations
ABMs and Social Network Studies: What and Why? As will be seen in the next few slides (What is ABM? ), there are several similarities of focus: • Dense interaction among the components of a social system (social embeddedness) • The behaviour of the whole as well as the parts (limited functional decomposability) • Formalisation of social organisation This suggests that techniques developed in one field may be applied to the other to bring insight in some studies
Background on Agent-based Modelling Background in Distributed Artificial Intelligence (DAI) Properties of Agents Properties of Agent Systems • Autonomy • Flexible • Adaptation • Scalable • Interactive • Distributed • Heterogeneous • Robust Many of these properties are shared with social systems argument for the usability of the approach
Basic principles • Agents represent the actors in the system, i. e. firms, institutions • We define agent characteristics as well as their behaviour • These are implemented as rules in the computer program • An agent is like an object in OOP …. • … but normally it has some goals, some means of perceiving its environment, and some kind of reasoning mechanism • Agents should be embedded within social context
Examples • Behavioural norms such as fashion trends or religion • Group behaviour such as in crowds, traffic or urban spaces • Environmental models of land use change or water resources • Consumer behaviour in retail markets • Auctions and supply-chain models Note that we have seen a quick overview of ABMs: more practical information on methodology of ABM will follow in the afternoon session
Software • Java Development Kit (JDK) version 1. 5. 0 – object orientated, platform independent, widely used. Arranged into packages. • Re. Past 3. 1 – Set of Java packages for ABM. GUI for visualisation. Bytecode in repast. jar • Real. J IDE – Simple environment for editing java files, compiling and running programs • PAJEK – network analysis software
Introducing the JDK and Real. J JDK consists of the bytecodes for the whole Java core, as well as the tools for compiling (javac. exe) and running (java. exe) your own Java programs Real. J is a text editor for working on Java projects which has some built-in functions for linking with the Java tools A. K. A. Integrated Development Environment (IDE) Real. J splits the workspace into three components: text editor, project window, console panel
Introducing Re. Past can be used to implement dynamic agent-based models that describe state changes in simulated time Re. Past is a set of Java packages, which incorporates a Graphical User Interface (GUI) for visualisation. It has packages for importing and exporting network data Bytecode (Java class files) are contained in repast. jar
Re. Past basics Re. Past divides model implementation into separate parts: Setup sets (or resets) any initial parameters to their defaults and sets any objects to ‘null’ Build. Model creates the representational parts of the simulation, i. e. , the agents and their environment Build. Display builds those parts of the simulation needed for graphically displaying the simulation Build. Schedule schedules ‘actions’ that change the simulation’s state i. e. , that describe dynamic simulation of social processes
Running the Re. Past Demos 1. Launch Re. Past (Repast. exe), Add and Load the model, and input your parameters in the Re. Past toolbar 2. To generate. net files you will need to fill in the following fields: 3. pajek. Interval – the number of ‘ticks’ between each recording 4. filename. Path – the location for saving net files 5. (user directory) + mydirectory/myrun + (tick number) 3. Press (Set up) and then (Run). (Pause) or (Stop) the simulation and investigate via the Re. Past display 4. Locate (in user directory/data) the output files for your
Demo Models - Jin. Gir. New. Net Jin, Girvan, and Newman working paper: “The Structure of Growing Social Networks” (1) meetings take place between pairs of individuals at a rate which is high if a pair has one or more mutual friends and low otherwise; (2) acquaintances between pairs of individuals who rarely meet decay over time; (3) there is an upper limit on the number of friendships an individual can maintain
Demo Models - Jin. Gir. New. Net RED (Random) and GREEN (Neighbour) links Characterisation of outcomes: (1) Initially the network rapidly increases in density due to the addition of random links (2) Eventually the network becomes more cliqueish or clustered due to the formation of neighbour links
Demo Models (2): Innovation Networks The agent is a Firm, located upon 2 d grid, and connected to neighbours in cardinal directions (N, S, E, W) within visible range From Epstein and Axtell, Growing Artificial Societies • • • Aims at innovating, either individually or in partnership with other firms Endowed with a ‘skill profile’ (SP) of possessed skills Involved in an ‘individual learning’ process to acquire new skills in the universe of firms’ skills
Agent develops SP through depth-first search 1 2 Advanced skills depend upon prior acquisition of more basic skills Specialisation and differentiation of each agent 3 4
Demo Models (2): Innovation Networks • Innovations are specified as a set of skills which can be combined to develop a new product or production process Assumption that innovating firms gain visibility: Their neighbourhood (which defines possible partners) increases in size The simulation cycle: 1. Firm’s individual learning step 2. Individual innovation step 3. Joint innovation step Re. Past displays initial neighbour (RED) and current partnership (WHITE) relations as well as the partnership history (BLUE)
- Slides: 15