| Hi-index | 0.00 |
Online social interactions are useful in information seekingfrom digital libraries, but how to measure their influence on theuser's information access actions has not yet been revealed.Studies on this problem give us interesting insights into theworkings of human dynamics in the context of information accessfrom digital libraries. On the basis, we wish to improve thetechnological supports to provide more intelligent services in theongoing China-America Million Books Digital Library so that it canreach its potential in serving human needs.Our research aims at developing a common framework to modelonline social interaction process in community-assisted digitallibraries. The underlying philosophy of our work is that the onlinesocial interaction can be viewed as a dynamic process, and the nextstate of each participant in this process (e.g., personalinformation access competency) depends on the value of the previousstates of all participants involving interactions in the period.Hence, considering the dynamics of interaction process, we modeleach participant with a Hidden Markov Model (HMM) chain and thenemploy the Influence Model, which was developed by C.Asavathiratham as a Dynamic Bayes Net (DBN) of representing theinfluences a number of Markov chains have on each other, to analyzethe effects of participants influencing each other. Therefore, onecan think of the entire interaction process as a DBN frameworkhaving two levels of structure: the local level and the networklevel. Each participant i has a local HMM chain&Ggr;(A) which characterizes the transition of hisinternal states in the interaction process with state-transitionprobability∑overjdijP(Sit|Sjt-1) (Here states are hispersonal information access competence in different periods, whileobservations are his information access actions). Meanwhile, thenetwork level, which is described by a network graph&Ggr;(DT) whereD={dij} is the influence factormatrix, represents the interacting relations betweenparticipants. The strength of each connection,dij, describes the influence factor of theparticipant j at its begin on the one i at its end.Hence, this model describes the dynamic inter-influence process ofthe internal states of all participants involving onlineinteractions.To automatically build the model, we need firstly to extractobserved features from the data of online social interactions andinformation access actions. Obviously, the effects of interactionsare stronger if messages are exchanged more frequently, or theparticipants access more information in the online digitallibraries during the period of time. Based on this consideration,we select the interaction measureIMi,jt and the amount ofinformation IAjt as the estimationfeatures of xit. The interactionmeasure IAit and the amount ofinformation parameterize the features calculated automatically fromthe data of online social interactions between the participantsi and j, and the features calculated from the data ofinformation access actions respectively. Secondly, we need todevelop a mechanism for learning the parametersdij andP(Sit|Sjt-1.Given sequences of observations {xit}for each chain i, we may easily utilize theExpectation-Maximization algorithm or the gradient-based learningalgorithm to get their estimation equations.We ran our experiments in the online digital library of W3CConsortium (www.w3c.org), which contains a mass of news, electronicpapers or other materials related to web technologies. Users mayaccess and download any information and materials in this digitallibrary, and also may free discuss on any related technologicalproblems by means of its mailing lists. Six users were selected inour experiments to collaboratively perform paper -gathering tasksrelated to four given topics. Any user might call for help from theothers through the mailing lists when had difficulties in thisprocess. All participants were required to record subjectiveevaluations of the effects that the others influenced his tasks.Each experiment was scheduled by ten phases. And in each phase, wesampled IMi,jt andIAit for each participant and then fedthem into the learning algorithms to automatically build theinfluence model. By comparing with the subjective influence graphs,the experimental results show that the influence model can estimateapproximately the influences of online social interactions.