Please use this identifier to cite or link to this item:
Title: Coping with subjectivity and dishonesty in opinion evaluation by exploiting social factors
Authors: Fang, Hui
Keywords: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Issue Date: 2014
Source: Fang, H. (2014). Coping with subjectivity and dishonesty in opinion evaluation by exploiting social factors. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: In large and open environments, users may often encounter entities (e.g. people, products and information) which they have no previous experience with or prior knowledge of. In this case, they usually rely on the experience or knowledge of other users (referred as advisors) in the form of opinions (e.g. reviews and ratings), to choose which entities to interact with. However, in these environments, users can freely express their opinions, and the quality of opinions may then vary. Of all the possible reasons that result in diversity of opinion quality, the following two are the most dominant: (1) users are subjectively different, which thus leads to discrepancy of users' opinions towards same entities; and (2) some users might be dishonest and lie about their experience with entities. However, some existing approaches may only consider either dishonesty or subjectivity difference, while others cannot accurately distinguish these two different aspects. On the other hand, subjectivity difference and dishonesty, regarded as either the extrinsic or intrinsic characteristics of users (human-beings), are more related to the behavioral and psychological perspectives of users other than merely the statistical probability of data points that prevalently adopted by Computer Science. In other words, it is expected to involve understanding users' motivation and needs of providing opinions when evaluating them. It is thus worthwhile to incorporate social factors when addressing these two aspects for opinion evaluation. With the aforementioned two issues in mind, we aim to solve the subjectivity and dishonesty problems for opinion evaluation by also taking social factors into consideration in this dissertation. In particular, this dissertation is comprised of four studies. First, we propose a novel trust model stemmed from diffusion theory in Social Science (called DiffTrust), to evaluate the opinions of advisors by modeling their trustworthiness. Trust has been recognized as a diffusive concept. When modeling trust, it is crucial to consider the processes through which trust is cultivated in a system. On the other side, diffusion theory in Social Science seeks to explain how, why, and at what rate a new innovation spreads through a community. It is thus natural to derive a trust model from this well-studied theory by considering an advisor's trustworthiness as an innovation. Her trustworthiness perceived by a specific user is influenced by: the advisor's characteristics directly observed by the user, susceptibility of the user, the contagious influence of other users already having a certain level of trust on the advisor, and information about the environment. DiffTrust can capture the dynamics of trust, and its dependency on other users and the environment. However, it cannot accurately distinguish subjectivity and dishonesty. Second, we design a subjectivity alignment approach for reputation computation (SARC) when aggregating numerical ratings (opinions) provided by users towards the same entities. Subjectivity difference between users may come from two sources by analyzing the scenario of a user providing a rating towards an entity, from both psychological and behavioral perspectives: (1) intra-attribute subjectivity, the subjectivity in evaluating the same attribute; and (2) extra-attribute subjectivity, the subjectivity in evaluating different attributes. We learn these two kinds of subjectivity for each user by applying Bayesian learning and regression analysis on the basis of each user's past experience, respectively. Ratings provided by one user can thus be aligned for another user according to the two users' subjectivity. Although SARC only considers the subjectivity difference aspect, it is not much affected by dishonest users, as validated in our experiments. Finally, we also propose two approaches by explicitly distinguishing subjectivity and dishonesty. In the first approach, we present a novel probabilistic graphical trust model (PGTM) to separately consider these two aspects. We model the factors of advisors' intrinsic nature (dishonesty, i.e. benevolence, integrity and competence from Social Science), users' propensity to trust advisors, and subjectivity difference between users and advisors, as latent variables in the model that may influence users' trust towards advisors. PGTM ignores the fact that dishonesty and subjectivity overlap with each other to certain extent. Therefore, we further propose a two-layered clustering approach (SubGroup) by modeling each advisor as part of groups. Specifically, on the basis of indicative features extracted from users' rating behavior to distinguish subjectivity and dishonesty, in the first layer, each user clusters her advisors into different subjectivity groups and dishonest types, with respect to their rating behavior. In the second layer, each advisor is assigned to two groups with respective membership degrees. An alignment approach is designed to help each user align advisors' ratings to the ones of her own. In summary, the main technical contributions of this dissertation are three-folds: (1) DiffTrust, stemmed from the diffusion theory in Social Science, mainly deals with advisors' dishonesty; (2) SARC addresses two kinds of subjectivity difference between users; and (3) PGTM and SubGroup approaches can well distinguish and cope with dishonesty and subjectivity difference for opinion evaluation. As a new attempt to consider social factors in trust assessment, our approaches contribute to bridging the research gap between computational trust in Computer Science and psychological and behavioral trust in Social Science. Further, we hope that these in-depth studies induce more attention towards this important interdisciplinary research direction.
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Theses

Files in This Item:
File Description SizeFormat 
thesis.pdfthesis1.4 MBAdobe PDFThumbnail

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.