Source-linked AI summary
In-Depth Behavior Understanding and Use: The Behavior Informatics Approach
Longbing Cao
TL;DR
Existing behavior analysis often relies on demographic and transactional data that does not explicitly represent behavioral elements or dynamic and impact-oriented aspects. The paper proposes Behavior Informatics (BI), which converts source data into behavioral data for systematic behavior analysis, and reports promising potential across two real-world case studies.
Problem
Existing behavior analysis commonly uses demographic and transactional data, while dynamic, sequential, social, and impact-oriented behavior aspects lack an explicit systematic view.
Method
Behavior Informatics converts transactional source data into explicit behavioral data and organizes behavior representation, construction, impact analysis, pattern analysis, simulation, presentation, and use.
Results
BI showed promising potential in real-world market-surveillance and social-security overpayment-prevention case studies, supporting deeper behavior-oriented understanding and solutions.
Takeaways & Limitations
BI can complement classic analytical approaches by supporting more comprehensive and in-depth understanding of behavior-oriented business problems.
Takeaways & Limitations
BI remains an open area with unresolved issues in behavior modeling, simulation, impact analysis, pattern analysis, and presentation.
Abstract
from arXiv · showhide
The in-depth analysis of human behavior has been increasingly recognized as a crucial means for disclosing interior driving forces, causes and impact on businesses in handling many challenging issues. The modeling and analysis of behaviors in virtual organizations is an open area. Traditional behavior modeling mainly relies on qualitative methods from behavioral science and social science perspectives. The so-called behavior analysis is actually based on human demographic and business usage data, where behavior-oriented elements are hidden in routinely collected transactional data. As a result, it is ineffective or even impossible to deeply scrutinize native behavior intention, lifecycle and impact on complex problems and business issues. We propose the approach of Behavior Informatics (BI), in order to support explicit and quantitative behavior involvement through a conversion from source data to behavioral data, and further conduct genuine analysis of behavior patterns and impacts. BI consists of key components including behavior representation, behavioral data construction, behavior impact analysis, behavior pattern analysis, behavior simulation, and behavior presentation and behavior use. We discuss the concepts of behavior and an abstract behavioral model, as well as the research tasks, process and theoretical underpinnings of BI. Substantial experiments have shown that BI has the potential to greatly complement the existing empirical and specific means by finding deeper and more informative patterns leading to greater in-depth behavior understanding. BI creates new directions and means to enhance the quantitative, formal and systematic modeling and analysis of behaviors in both physical and virtual organizations.
1. Introduction
Traditional behavior analysis often infers behavior from demographic and transactional data, leaving behavioral elements, intentions, and impacts implicit. Behavior Informatics (BI) converts source data into explicit behavioral data for quantitative pattern and impact analysis, illustrated through business case studies.
- Traditional analyses commonly use customer demographics and transactions, although these records do not directly represent genuine behavioral elements.Examples include telecom churn classification from subscriber histories and abnormal-trading detection from price movement.
- Because behavioral properties are dispersed across transactions, analyses tend to focus on business appearances rather than behavior interior, intention, or impact.The paper identifies this implication as limiting in-depth scrutiny of behavioral causes and effects.
- The authors report that BI can complement existing analytical approaches by revealing deeper behavior patterns and supporting more comprehensive problem-solving.The introduction frames behavioral data as an additional perspective alongside demographic and transactional analysis.
- BI converts transactional and business-management data into behavioral data by extracting, transforming, presenting, and reorganizing hidden behavioral elements.The resulting data organizes behavior through entities, properties, and relationships.
- BI includes behavior representation, behavioral data construction, impact and pattern analysis, simulation, presentation, and behavior use.The framework is positioned primarily from information-technology and data-analysis perspectives.
- The paper illustrates BI with customer churn analysis and case studies of market microstructure and social-security activity patterns.Experiments use real-world stock-market and Australian government data to examine critical application problems.
2. Related Work
Related work spans user modeling, activity monitoring, customer and web behavior analysis, contextual techniques, sequence analysis, and reality mining. The paper identifies a gap in systematic analysis of dynamic, sequential, social, and impact-oriented behavior.
- User modeling in human-computer interaction emphasizes cognitive models, while machine-learning approaches often predict future user actions.These objectives differ from the paper’s broader behavior-modeling focus.
- Activity monitoring targets unusual behavior, and web-usage research uses access, weblog, and session data to analyze navigational history, experience, and location.Online-business studies also examine product awareness and purchase commitment for pattern discovery.
- Customer and consumer behavior research addresses marketing, customer relationships, social factors in churn, fraud, and insider trading using relational or price information.The cited applications show varied domain-specific approaches to behavior-related problems.
- Context representation, ontological engineering, semantic web methods, sequence analysis, and reality mining provide additional tools relevant to behavior analysis.These methods respectively contribute contextual, semantic, sequential, or machine-sensed environmental information.
- Existing behavior analysis often emphasizes static, separately studied properties and business appearances or unusual events rather than an explicit systematic view.The paper identifies dynamic, sequential, social, and impact-oriented analysis as underrepresented dimensions.
- Activity mining extends prior work through activity-data modeling, pattern analysis, and impact analysis, including applications to exceptional behavior and market trading.The paper situates BI partly on these preliminary studies.
3. An Empirical Behavioral Model
The paper defines BI behavior as computationally recorded actions, operations, events, or activity sequences and represents it through domain-adaptable behavioral vectors. These vectors support sequence-based pattern analysis beyond traditional sequential mining.
- What Is Behavior About?: BI focuses on computationally recorded or converted behaviors, including actions, operations, events, and activity sequences within organizational contexts.The framework distinguishes individual and group behaviors and symbolic from mapped behavior.
- What Is Behavior About?: Symbolic behavior represents social activities recorded as symbols, whereas mapped behavior represents physical activities captured by sensors.Stock-trading actions exemplify symbolic behavior; video surveillance exemplifies mapped behavior.
- What Is Behavior About?: For mapped behavior, BI focuses on pattern analysis after detection and representation rather than on detecting the underlying physical behavior.Detected behaviors are extracted into behavioral data similarly to symbolic behaviors.
- An Empirical Behavioral Model: A behavioral vector represents a subject acting on an object through attributes including context, goal, belief, action, plan, impact, constraint, time, place, status, and associates.The model captures both basic properties and social or organizational factors.
- An Empirical Behavioral Model: The model defines action as what the subject chooses to do, plan as action sequences supporting intentions, and impact as results produced on an object or context.Constraint, time, place, status, and associates describe conditions, occurrence, location, stage, and related behaviors.
- An Empirical Behavioral Model: The behavioral vector is γ⃗ = {s, o, e, g, b, a, l, f, c, t, w, u, m}, and customer behavior sequences comprise sequences of such vectors.The vector may contain textual, categorical, and numerical data.
- An Empirical Behavioral Model: Deployments may omit, specialize, or expand attributes, and vector-based analysis is intended to identify richer patterns than traditional sequential pattern mining.The paper states that existing mining techniques cannot be directly applied to this complex data structure.
4. Framework of Behavior Informatics
Behavior Informatics (BI) provides a multidisciplinary framework for converting transactional data into behavior-oriented representations, then analyzing, simulating, presenting, and using behavioral patterns and impacts.
- Research scope: BI addresses behavioral data construction, representation, impact modeling, pattern analysis, simulation, presentation, and use as interconnected research tasks.The framework also identifies measurement and evaluation as important topics and treats behavior simulation as relevant to artificial systems and real societies.
- Research scope: Behavioral data construction converts normal source data into a behavior-oriented feature space through feature selection, mapping, transformation, and quality control.The goal is for behavior elements to constitute the major proportion of the resulting dataset.
- Research scope: Behavior modeling and representation develops formal languages and tools to capture behavioral entities, properties, relationships, interactions, causality, evolution, and emergence.These techniques support understanding behavior entities, behavior networks, and behavior impacts.
- Research scope: Behavior pattern and impact analysis examine behavior entities, networks, structures, dynamics, risks, costs, trust, and effects across economic, cultural, organizational, social, and political settings.Pattern analysis includes detection, prediction, and prevention of critical behavior, misbehavior, and behavior impact.
- Process: The generic BI process converts entity-relationship transactional data into behavior-feature data, analyzes patterns and impacts, presents results, and transforms them into decision-support business rules.Behavior simulation and modeling can provide foundational results about behavior dynamics and relevant business contexts for knowledge discovery.
- Theoretical underpinnings: BI draws on analytical, computational, and social sciences, combining technologies such as formal methods, ontologies, machine learning, simulation, multiagent systems, and social network analysis.Operational techniques include algebra, logics, classification, sequence analysis, and tools for representing, modeling, analyzing, presenting, and using behavior.
5. Case Study 1: Market Microstructure Behavior Analysis
The case study converts market microstructure transactions into explicit behavioral representations that capture investors’ actions, intentions, lifecycles, and impacts. It then mines technically and commercially interesting behavior patterns to support deeper market analysis and surveillance.
- Market Microstructure Behavior in Capital Markets: Market microstructure transactions record orders, trades, indices, and market data, but primarily represent the results of investors’ behavior rather than their intentions.These data involve dimensions such as time, value, and trade actions, while behavior-related properties remain implicit.
- Market Microstructure Behavior in Capital Markets: Behavior modeling extracts behavioral elements from orderbook transactions and represents them as explicit microstructure behavioral data.This conversion shifts behavior from transactional space into behavioral data suitable for behavior-oriented analysis.
- Modeling Market Microstructure Behavior: The microstructure behavior vector represents an investor, security, price, volume, trading action, trading probability, impact, trading status, and follow-up actions.The model maps transaction attributes to behavioral concepts while retaining the relations needed to describe trading activity.
- Modeling Market Microstructure Behavior: Vector-based behavioral sequences encode investors’ intentions, trading activities, associated behavior instances, and behavior procedures or lifecycles within a trading period.Investor data such as account identifiers are used to construct sequences from behavior-related orderbook transactions.
- Mining Microstructure Behavior Patterns: Exceptional behavior patterns are selected when they satisfy both technical and business-interestingness criteria, producing actionable patterns.Thresholds are defined and refined using domain expertise and experimental findings.
- Experiments: Compared with price-centered analysis, microstructure behavior patterns provide deeper information about interior trading activities, processes, resulting status, and investment impact.The authors state that this information can assist market surveillance officers in understanding price movement and related investors’ driving forces.
6. Case Study 2: Social Security Behavior Analysis
The social security case study constructs behavioral data by combining customer demographics with coded interaction and intervention activity sequences. It then mines debt-related patterns to identify risk indicators, contrasting impacts, and demographic conditions associated with different repayment outcomes.
- Modeling and constructing activity sequences: Debt-related activities are extracted from generic transactions and recoded into activity codes representing customer–officer interactions and intervention actions.This produces activity data corresponding to customer interactions with the government.
- Behavioral data construction: The behavioral representation separates customer demographics into one vector and interaction or intervention activities into another, combining both for analysis.The combined representation is intended to provide a more complete understanding of debt causes and effects than demographic-only data.
- Modeling and constructing activity sequences: Activity sequences contain coded activities within a customer-specific time window, followed by an impact label distinguishing debt (DET) from non-debt (NDT).The examples include a debt sequence with a $315 debt lasting 14 days and a non-debt sequence.
- Behavioral data construction: The constructed demographic and activity vectors form social-security behavioral data for subsequent pattern and impact analysis.The resulting representation is Γ = {Γ1, Γ2}.
- Impact-reversed activity patterns: Impact-reversed pattern pairs identify sequences that can convert behavioral impact from negative to positive through inducing or blocking a consequential Q-sequence.The paper presents these patterns as useful for debt-prevention action-taking.
- Demographic-activity-combined patterns: Combined demographic, arrangement, and repayment patterns distinguish customers whose similar arrangements and repayment actions produce different payback effects under different demographic circumstances.The resulting clusters provide indicators for targeting actions that may convert slow payers into moderate or quick payers.
7. Conclusions
The paper proposes Behavior Informatics (BI) for explicit, in-depth analysis of genuine behavior in business problems, addressing behavioral information hidden in transactional data. Case studies suggest BI can complement classic analytical approaches, while its framework and techniques require further development and evaluation.
- Behavior is implicitly and separately recorded in transactional data, limiting analysis based mainly on demographic and service-usage information.The paper frames this as a barrier to understanding genuine behavioral actions, operations, events, and their roles in business problems.
- Behavior Informatics (BI) provides a research framework for explicit and in-depth behavior analysis, including behavior modeling, pattern analysis, simulation, impact analysis, presentation, and use.The framework starts from a definition of behavior and a behavioral model and covers the associated research issues, processes, and technical underpinnings.
- Experiments in capital markets and social security indicate that BI is important, feasible, and effective for understanding behavior-oriented problems and presenting behavior-oriented solutions.The paper reports two real-world case studies: market microstructure behavior analysis and social-security behavior analysis.
- BI has the potential to complement classic analytical approaches by supporting more comprehensive and in-depth business understanding and problem-solving.This conclusion is stated within the scope of the reported case studies and experiments.
- Further studies are needed to assess the completeness, sufficiency, and effectiveness of the proposed framework and potential solutions.The paper characterizes BI as a newly addressed field whose complexity warrants continued investigation through concrete real-life case studies.