Elsevier

Information Sciences

Volume 252, 10 December 2013, Pages 1-19
Information Sciences

Behavioral modeling with the new bio-inspired coordination generalized molecule model algorithm

https://doi.org/10.1016/j.ins.2011.12.003Get rights and content

Abstract

Social Networks (SN) is an increasingly popular topic in artificial intelligence research. One of the key directions is to model and study the behaviors of social agents. In this paper, we propose a new computational model which can serve as a powerful tool for the analysis of SN. Specifically, we add to the traditional sociometric methods a novel analytical method in order to deal with social behaviors more effectively, and then present a new bio-inspired model, the coordination generalized molecule model (CGMM). The proposed analytical method for social behaviors and CGMM are combined to give an algorithm that can be used to solve complex problems in SN. Traditionally, SN models were mainly descriptive and were built at a very coarse level, typically with only a few global parameters, and turned out to be not sufficiently useful for analyzing social behaviors. In this work, we explore bio-inspired analytical models for analyzing social behaviors of intelligent agents. Our objective is to propose an effective and practical method to model intelligent systems and their behaviors in an open and complex unpredictable world.

Introduction

Social networks and social behaviors as research topics have attracted much attention in many disciplines including modern sociology, anthropology, sociolinguistics, geography, social psychology, communication studies, information science, organizational studies, economics and biology. In computer science and engineering, there is research on social networks dealing with such aspects as intelligent behaviors, and learning and adaptation in a social context by machines.

Social behavior can be seen as diametric to individual behavior from the perspective of AI and philosophy [2]. Intelligent systems that are oriented towards some goals would exhibit certain kind of social behaviors [5]. For them, we can define social behavior as goal-oriented behavior. In this paper, we only discuss and analyze goal-oriented behaviors, goal-oriented agents and goal-oriented systems. Note that goal-oriented systems may or may not maintain any explicit internal representation of their goals. The notion of goal-oriented behavior is based on the operational notion of goal and “purposive behavior” as proposed by Rosenblueth and Wiener [18], and has been further developed, in psychology, by Miller, Galanter and Pribram [15].

Here is a quick summary of the related works on social behavior: (1) Malrey Lee translates the artificial life method in emergent behavior evolution of autonomous mobile robots into a modeling of the pursuit system with artificial neural networks and genetic algorithms [12]. (2) In [9], Hervé Frezza-Buet and Frédéric Alexandre present a cortical model, which is inspired by biological data, to allow an animat to perceive and act on its environment to learn the regularities of the environment and to use them to satisfy certain goals; they then describe this model at different levels of abstraction, from both a biological and a computational point of view. (3) Julio Rosenblatt et al. develop a behavior-based system for controlling an autonomous underwater vehicle to perform a survey of coral reefs [17]; the implemented behaviors have the ability to avoid collisions, to maintain a proper standoff distance, and to follow the transect using video or sonar. (4) In [13], Seungho Lee et al. extend the decision field theory (DFT) for behavior modeling in a dynamic environment using Bayesian belief networks. (5) Peeva and Zahariev define finite fuzzy machines and investigate their behaviors [16]; algorithms are proposed for computing behaviors, for establishing equivalence and redundancy of states, and for solving reduction and minimization problems. (6) Yi-Chun Chang and Chih-Ping Chu propose the Learning Behavioral Petri Nets (LBPN) to model learning behavior in web-based environments [4]. (7) In [14], James Liou and Gwo-Hshiung Tzeng use the Dominance-based Rough Set Approach (DRSA) to provide a set of rules for determining customer attitudes and loyalties, which can help managers develop strategies to acquire new customers and retain highly valued ones. (8) Frank Hoffmann presents a soft computing approach for robot behavior design, which combines evolutionary algorithms, fuzzy control and neural networks [7]. (9) In [22], Peter Walley and Gert de Cooman show that it is possible to formulate behavioral models for representing linguistic uncertainties in terms of upper probability measures.

This paper has two main contributions. First, it describes a method making use of categorization, quantization and matrix representation for analyzing social behaviors in social networks (SN). Second, it proposes a new bio-inspired model—the coordination generalized molecule model (CGMM)—and its algorithm, which can be combined with the above analytical method to solve complex problems in SN.

Conventional social networks are known as symbolic SN, logical SN, semantic SN, and so on. We develop computational SN, which is iterative and its learning is based on non-symbolic methods and bio-inspired computing. In this paper, we extend traditional sociometric methods using a novel analytical method in order to better deal with social behaviors. We study an SN using categorization, quantization, mathematical (matrix) representation, and test and verification, with the hope that the resulting bio-inspired analytical method can accurately mimic the evolution of social networks.

We present a CGMM algorithm for SN, which is based on the GP algorithm [6] and the proposed analytical method of social behaviors. In fact, CGMM is motivated by observations of various social behaviors and the kinematics and dynamics of molecules in biology. The latter exhibit parallelism, openness, local interactivity, and self-organization. We extend the biological molecule model to our coordination generalized molecule model (CGMM) for treating social behaviors in SN. The following summarizes the differences of the CGMM algorithm from the GP algorithm: (1) the CGMM algorithm aims at studying social behaviors in SN, and can deal with complex behaviors and dynamics; (2) there is a “force-field” in the CGMM architecture; (3) in CGMM, each agent in the SN is transformed into a molecule in the force-field; (4) the proposed analytical method for social behaviors (comprising categorization, quantization, and mathematical representation of social behaviors) is integrated into the CGMM algorithm to solve complex problems in SN.

The CGMM algorithm can overcome some of the limitations of existing popular solutions for SN: (1) it can describe complex behaviors and dynamics; (2) it to perform large-scale distributed optimization; (3) it can converge; (4) its setting of parameters does not depend on algorithmic learning or any apriori knowledge; (5) it has a comprehensive optimization ability for multiple objectives; (5) it is robust—CGMM is basically independent of the initial conditions, problem size, small-range parameter changes, etc.; (7) it has a powerful processing ability in a complex and dynamically changing environment; (8) it has greater flexibility and is very easy to adapt to a wide range of optimization problems. All in all, it is fundamentally different from the other popular approaches for SN in its motivation, the principle behind, its optimization mechanism, its elements and their states, and the biological, and the mathematical and theoretical models on which it is founded.

This paper has six sections. We categorize social behaviors, quantify them, propose a mathematical (matrix) representation for them, and test and verify the proposed CGMM approach in Sections 2 Categorization, 3 Quantification, 4 SN matrix representation, 5 Test and verification: the CGMM algorithm respectively. Conclusions are drawn in Section 6.

Section snippets

Categorization

Fig. 1 shows an example of a social network. Sociality requires that there are at least two or more agents in a social network. In a social network, there are interferences among the behaviors of the agents as they all try to achieve their goals.

In general, the behaviors of one agent produce effects that are relevant for the goals of another. Interference can be divided into positive interference and negative interference [3], [8].

  • Positive interference: the behavior of one agent helps the

Quantification

Of the 12 types of social coordination, types AF are unilateral coordination, and types GL are bilateral communication. Based on which agent (s) will modify their current intention, the 12 types can be conveniently grouped into the four categories. For Aijk, Bijk, Cijk, it is Ai that will modify its current intention; for Dijk, Eijk,Fijk, it is Aj; for Gijk, Hijk, Iijk, it is both Ai and Aj; and for Jijk, Kijk, Lijk, none will; so we haveL(I)=L(10)=Aijk,Bijk,Cijk|i,j,k,L(II)=L(01)=Dijk,Eijk,F

SN matrix representation

In the simplest social network, every agent interacts with every other agent. This means that the underlying topology is a graph with all-to-all connectivity. But for different objects, the agents’ social behaviors might be different. We give an example of a robot, which in fact is a very simple social network, to illustrate the many kinds of social behaviors among the agents. We transform this example social network into its topology (node-link diagram) and a matrix representation.

The robot

Test and verification: the CGMM algorithm

Without loss of generality, we choose and focus on an optimization problem involving social behavior in a social network to demonstrate how to deal with social behaviors. We combine the social behavior analysis method mentioned above and a bio-inspired Coordination Generalized Molecule Model (CGMM) algorithm to solve the optimization problem.

Conclusion

In this paper, we propose a social behavior analysis method which comprises categorization, quantization, and matrix representation. We also present a new bio-inspired model, the coordination generalized molecule model (CGMM) and its algorithm. We combine the behavior analysis method and the CGMM algorithm to solve optimization problems due to social behaviors in social networks. The particular optimization problem we have solved successfully confirms the effectiveness of our social behavior

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China under Grant Nos. 60905043, 61073107 and 61173048, the General Research Fund of Hong Kong Research Grant Council under Grant No. 7137/08E, the Innovation Program of Shanghai Municipal Education Commission, and the Fundamental Research Funds for the Central Universities.

Cited by (2)

  • Solving partially observable problems with inaccurate PSR models

    2014, Information Sciences
    Citation Excerpt :

    One of the most challenging problems in artificial intelligence (AI) is concerned with agents operating in environments that are partially observable and stochastic [9,28,30], i.e., how an agent can plan and act optimally in such environments. Commonly used techniques model the system firstly and then solve the problem using the obtained model [8,24]. Thus far, the most general framework to model such controlled dynamical systems is partially observable Markov decision processes (POMDPs) [22].

  • The anti-bouncing data stream model for web usage streams with intralinkings

    2014, Information Sciences
    Citation Excerpt :

    It gives a more authentic summarization of user behaviors and a more reliable measurement of bounce rate. In this work, we consider the case of bounce rate and clustering, but ABS should have the same advantages for other techniques such as pattern extraction [7,13], behavioural modeling [6,26,15], recommendation [40] or stream analysis in social networks [30]. In this paper, we extend the work of [48] and elaborate in depth about the data-intralinking problem in Web usage streams and the solutions, including detailed techniques and experimental results.

View full text