Behavior Computing

The popularity of social media as an information source, in the recent years has spawned several interesting applications, and consequently challenges to using it effectively. Identifying and targeting influential individuals on sites is a crucial way to maximize the returns of advertising and marketing efforts. Recently, this problem has been well studied in the context of blogs, microblogs, and other forms of social media sites. Understanding how these users behave on a social media site and even across social media sites will lead to more effective strategies. In this book chapter, we present existing techniques to identify influential individuals in a social media site. We present a user identification strategy, which can help us to identify influential individuals across sites. Using a combination of these approaches we present a study of the characteristics and behavior of influential individuals across sites. We evaluate our approaches on several of the popular social media sites. Among other interesting findings, we discover that influential individuals on one site are more likely to be influential on other sites as well. We also find that influential users are more likely to connect to other influential individuals.

Behavior modeling has been increasingly recognized as a crucial means for disclosing interior driving forces and impact in social activity processes. Traditional behavior modeling in behavior and social sciences that mainly relies on qualitative methods is not aimed at deep and quantitative analysis of social activities. However, with the booming needs of understanding customer behaviors and social networks etc., there is a shortage of formal, systematic and unified behavior modeling and analysis methodologies and techniques. This paper proposes a novel and unified general framework, called Social Activity Process Modeling and Analysis System (SAPMAS). Our approach is to model social behaviors and analyze social activity processes by using model checking. More specifically, we construct behavior models from sub-models of actor, action, environment and relationship, followed by the translation from concrete properties to formal temporal logic formulae, finally obtain analyzing results with model checker SPIN. Online shopping process is illustrated to explain this whole framework.

This chapter illustrates some of the different knowledge representation and inference tools used by a high-level, fully implemented conceptual language, NKRL (Narrative Knowledge Representation Language), to deal with the most common types of human “behaviours”. All possible kinds of multimedia “narratives”, fictional or non-fictional, can be seen in fact as streams of elementary events that concern the behaviours, in the most general meaning of this term, of some specific characters. These try to attain a specific result, experience particular situations, manipulate some (concrete or abstract) materials, send or receive messages, buy, sell, deliver, etc. Being able to deal in a correct (and computer-usable) way with narratives implies then being able to deal correctly with the behaviours of the concerned characters.

Semi-Markov models have been used in the recent past to model user navigation behavior for personalization in the field of Web-based applications. However, research on its application to incorporate personalization in generalized software packages is rare. In this paper, we use a semi-Markov model to dynamically display personalized information in the form of high-utility software functions (states) of a software package to a user. We develop a demo package of ActiveX Servers and Controls as a test-bed.

As the increasing popularity of social networking functions, people interact with others in social events everyday. However, people are easily overwhelmed by hundreds of social events. In this work, we propose P-SERS, a Personalized Social Event Recommender System, which consists of three phases: (1) Candidate selection, (2) Social measurement and (3) Recommendation. Among these, potential candidate events are selected based on user preference and the social network. In our opinion, every social event is composed of three critical elements: (1) the initiator, (2) the participants and (3) the target item. These elements possess different types of influential power on a social event. Therefore, we design algorithms to compute three social measures, i.e., initiator score, participant score and target score, which model expertise of the initiator, group influence of participants and global popularity of the target item respectively. P-SERS evaluates each candidate social event by these social measures and produces a recommendation list. In addition, explanations and the grouping function are provided to improve the recommendation. Finally, we examine P-SERS by recommending group buying events in a real world online group buying website. The experimental results show the superiority of P-SERS over conventional social recommendation methods.

Capturing individual profiles is one of the key tasks in behavior computing. Now, web forum has been one of the main platforms to exchange information. In this paper, we focus on get extension profiles for web forum users. Extension profiles are the types and areas of that forum users concern about (“term-profile”) plus a description of user’s collaboration networks (neighborhood-profile). We define and implement the tasks of automatically determining an extension profiles of a forum user from a web forum corpus. We propose the tripartite graph model to effectively capture the user’s profiles. This tripartite graph integrate forum user’s posts and reply networks in web forums. Furthermore, we discuss how to implement an application by following the model. And the efficiency of the application is discussed on the basis of an experimental study using a real data set of online forum.

Mining the behavioral characteristics and the adaptive learning mechanism of users during their information searching is meaningful for the academic database providers to improve their service and build e-learning platform to help users manipulate their search products more effectively The paper comprises four main parts: the first, Related work, makes a literature review and declares the concerns of our research; the second, Theories and models, explains the basic idea of reinforcement learning behavior, and introduces three representative reinforcement learning models, i.e. BM model, BS model and CR model; the third, Experiments and analysis, experimentally observes the characteristics of the academic users’ reinforcement learning in the process of search tasks performing, further quantitatively simulates their reinforcement learning behavior in information seeking using the three learning models, and gives extensive discussions about these models; and the fourth, Conclusions, makes some suggestions for the academic database providers efficiently. Based on the theories and models of reinforcement learning behavior, this research takes the freshmen and senior students from universities as user samples, experimentally observes the explicit behavioral and implicit psychological characteristics of their learning behavior in the process of search tasks performing, and further quantitatively simulates their reinforcement learning behavior in information seeking using the Bush-Mosteller model, Borgers-Sarinare model and Cross model. Finally, the paper makes some extensive discussions about these models and gives some advices to the database providers.

The ability to perceive similarities and group entities into meaningful hierarchies is central to the processes of learning and generalisation. In artificial intelligence and data mining, the similarity of symbolic data has been estimated by techniques ranging from feature-matching and correlation analysis to Latent Semantic Analysis (LSA). One set of techniques that has received very little attention are those based upon cognitive models of similarity and concept formation. In this paper, we propose an extension to a neural network-based approach called Forming Global Representations with Extended backPropagation (FGREP), and show that it can be used to form meaningful conceptual clusters from information about an entity’s perceivable attributes or its usage and interactions. By examining these clusters, and their classification errors, we also show that the groupings identified by FGREP are more intuitive, and generalise better, than those formed using LSA.

This study presents a methodology to determine risk scores of individuals, for a given financial risk preference survey. To this end, we use a regression-based iterative algorithm to determine the weights for survey questions in the scoring process. Next, we generate classification models to classify individuals into risk-averse and risk-seeking categories, using a subset of survey questions. We illustrate the methodology through a sample survey with 656 respondents. We find that the demographic (indirect) questions can be almost as successful as risk-related (direct) questions in predicting risk preference classes of respondents. Using a decision-tree based classification model, we discuss how one can generate actionable business rules based on the findings.

Implicit relevance feedback has received wide attention recently, as a means to capture the search context in improving search accuracy. However, implicit feedback is usually not available to public or even research communities at large for reasons like being a potential threat to privacy of web users. This makes it difficult to experiment and evaluate web search related research and especially web search personalization algorithms. Given these problems, we are motivated towards an artificial way of creating user relevance feedback, based on insights from query log analysis. We call this simulated feedback. We believe that simulated feedback can be immensely beneficial to web search engine and personalization research communities by greatly reducing efforts involved in collecting user feedback. The benefits from “Simulated feedback” are—It is easy to obtain and also the process of obtaining the feedback data is repeatable and customizable. In this chapter, we describe a simple yet effective approach for creating simulated feedback. We evaluated our system using clickthrough data of a set of real world users and achieved 65% accuracy in generating click-through data of those users.

In this chapter, we present a new trajectory pattern mining framework, namely, Clustering Clues of Trajectories (CCT), for discovering trajectory routes that represent frequent movement behaviors of a user. In addition to spatial and temporal biases, we observe that trajectories contain silent durations, i.e., the time durations when no data points are available to describe movements of users, which bring many challenge issues in clustering trajectories. We claim that a movement behavior would leave some clues in its various sampled/observed trajectories. These clues may be extracted from spatially and temporally co-located data points from the observed trajectories. Based on this observation, we propose clue-aware trajectory similarity to measure the clues between two trajectories. Accordingly, we further propose the clue-aware trajectory clustering algorithm to cluster similar trajectories into groups to capture the movement behaviors of the user. We validate our ideas and evaluate the proposed CCT framework by experiments using both synthetic and real datasets. Experimental results show that CCT is more effective in discovering trajectory patterns than the state-of-the-art techniques in trajectory clustering.

A fundamental challenge in behavioral informatics is the development of methodologies and systems that can achieve its goals and tasks, including behavior pattern analysis. This study presents such a methodology, that can be converted into a decision support system, by the appropriate integration of existing tools for association mining and graph visualization. The methodology enables the linking of behavioral patterns to personal attributes, through the re-mining of colored association graphs that represent item associations. The methodology is described and mathematically formalized, and is demonstrated in a case study related with retail industry.

Causality can be detectable from categorical data: hot weather causes dehydration, smoking causes cough, etc. However, in the context of numerical data, most of the times causality is difficult to detect and measure. In fact, considering two time series, although it is possible to measure the correlation between both associated variables, correlation metrics don’t show the cause-effect direction and then, cause and effect variables are not identified by those metrics.

In order to detect possible cause-effect relationships as well as measuring the strength of causality from non-categorical numerical data, this paper presents an approach which is a simple and efficient alternative to other methods based on regression models.

Frequent itemset mining generates frequently purchased itemsets, which only considers the presence of an item in a transaction database. However, a frequent itemset may not be the itemset with high value. High utility itemset mining considers both of the profits and purchased quantities for the items, which is to find the itemsets with high utility for the business. The previous approaches for mining high utility itemsets first apply frequent itemset mining algorithm to find candidate high utility itemsets, and then scan the whole database to compute the utilities of these candidates. However, these approaches need to take a lot of time to generate all the candidate high utility itemsets, scan the whole database and search from a large number of candidate high utility itemsets to compute the utilities of these candidates. Therefore, the previous approaches are very inefficient.

In this paper, we present an efficient algorithm for mining high utility itemsets. Our algorithm is based on a tree structure in which a part of utilities for the items are recorded. A mechanism is proposed to reduce the mining space and make our algorithm can directly generate high utility itemsets from the tree structure without candidate generation. The experimental results also show that our algorithm significantly outperforms the previous approaches.

Secure physical environments such as government, financial or military facilities are vulnerable to misuse by authorized users. To protect against potentially suspicious actions, data about the movement of users can be captured through the use of RFID tags and sensors, and patterns of suspicious behaviour detected in the captured data. This chapter presents four types of suspicious behavioural patterns, namely temporal, repetitive, displacement and out-of-sequence patterns, that may be observed in such a secure physical environment. We model the physical environment and apply algorithms for the detection of suspicious patterns to logs of RFID access data. Finally we present the design and implementation of an integrated system which uses our algorithms to detect suspicious behavioural patterns.

There are many issues concerning information security of digital libraries. Apart from traditional information security problems there are some specific ones for digital libraries. In this work we consider a behavioral modeling approach to discover unauthorized copying of a large amount of documents from a digital library. Supposing the regular user has interest in semantically related documents, we treat referencing to semantically unrelated documents as anomalous behavior that may indicate attempt of unauthorized large-scale copying. We use an adapted anomaly detection approach to discover attempts of unauthorized large-scale documents copying. We propose a method for constructing classifiers and profiles of regular users’ behavior based on application of Markov chains. We also present the results of experiments conducted within development of a prototype digital library protection system. Finally, examples of a normal profile and an automatically detected anomalous session derived from the real data logs of a digital library illustrate the suggested approach to the problem.

Everyday online communities and social networks are accessed by millions of Web users, who produce a huge amount of user-generated content (UGC). The UGC and its publication context typically evolve over time and reflect the actual user interests and behaviors. Thus, the application of data mining techniques to discover the evolution of common user behaviors and topic trends is becoming an appealing research issue. Dynamic association rule mining is a well-established technique to discover correlations, among data collected in consecutive time periods, whose main quality indexes (e.g., support and confidence) exceed a given threshold and possibly vary from one time period to another.

This Chapter presents the DyCoM (Dynamic Context Miner) data mining system. It entails the discovery of a novel and extended version of dynamic association rules, namely the dynamic generalized association rules, from both the content and the contextual features of the user-generated messages posted on Twitter. A taxonomy over contextual data features is semi-automatically built and exploited to discover dynamic correlations among data at different abstraction levels and their temporal evolution in a sequence of tweet collections.

Experiments, performed on both real Twitter posts and synthetic datasets, show the effectiveness and the efficiency of the proposed DyCoM framework in supporting user behavior and topic trend analysis from Twitter.

In Mobile Social Network Analysis, mobile users interaction pattern change frequently and hence it is very hard to detect their changing patterns because humans posses an extremely high degree of randomness in their calling behavior. To identify regularity in such random behavior, we propose a new method using network attributes to find periodic or near periodic graphs in dynamic social networks. We try to analyze real-world mobile social networks and extract its periodicity through a simple practical and efficient method using effective network attributes of the social network. We demonstrate the applicability of our approach on real-world networks and extract meaningful and interesting periodic interaction patterns. This helps in defining targeted business models in cellular communication arena.

In this paper we propose the model of the inhomogeneous Poisson for call frequency and inhomogeneous exponential distribution for call durations to detect events based on mobile phone call detail records. The maximum likelihood method is used to estimate the rate of frequency and call duration. This work is useful for enhancing homeland security, detecting unwanted calls (e.g., spam) and commercial purposes. For validation of our results, we used actual call logs of 100 users collected at MIT by the Reality Mining Project group for a period of 8 months. The experimental results show that our model achieves good performance with high accuracy.

Prediction of incoming calls can be useful in many applications such as social networks, (personal, business) calendar and avoiding voice spam. Predicting incoming calls using just the context is a challenging task. We believe that this is a new area of research in context-aware ambient intelligence. In this paper, we propose a call prediction scheme and investigate prediction based on callers’ behavior and history. We present Holt-Winters method to predict calls from frequent and periodic callers. The Holt-Winters method shows high accuracy. Prediction and efficient scheduling of calls can improve the security, productivity and ultimately the quality of life.

Based on accelerometer, we propose a 3D handwriting recognition system in this paper. The system is consists of 4 main parts: (1) data collection: a single tri-axis accelerometer is mounted on a handheld device to collect different handwriting data. A set of key patterns have to be written using the handheld device several times for consequential processing and training. (2) Data preprocessing: time series are mapped into eight octant of three-dimensional Euclidean coordinate system. (3) Data training: hidden Markov models (HMMs) and Gaussian mixture models (GMMs) are combined to perform the classification task. (4) Pattern recognition: using the trained HMM to carry out the prediction task. To evaluate the performance of our handwriting recognition model, we choose the experiment of recognizing a set of English words. The accuracy of classification could be achieved at about 96.5%.

Medical information searching has become an integral part of a medical students’ life. Yet, medical students are not provided with formal education on how to search for medical information on medical domains. Searching for medical information on medical domains is not a trivial task. It requires usage of appropriate terminology and the ability to comprehend returned results. The search behavior of medical students is rarely studied to understand how their information search goal is being satisfied or to identify specific search challenges faced by them. In this paper, we study interactive information searching behavior of medical students. Using simulated work task scenarios, we identify similarities and variability’s in search patterns and analyzes search behavior traits demonstrated by medicals students. Based on our findings, we suggest intuitive methods medical search engines could adapt to improve medical students’ search interaction to better support the search process.

This paper briefly surveys the software testing techniques based on the works of classification and evaluation. In addressing the two major software testing issues, that is when should testing stop and how good the technique is after testing, I present a scheme by a data flow diagram for evaluating software testing techniques. Following this diagram step by step, all the activities involved and the relative techniques were described. Software testing has progressed through five major paradigms they are the debugging, demonstration, destruction, evaluation and prevention periods, as outlined by a number of authors. During its development, software testing has focused on two separate issues, verification and validation. A strategy proposal for software testing in the development of applications is advocated later.