Chung-Wei Hang

AAAI Conference 2022 Conference Paper

Improved Text Classification via Contrastive Adversarial Training

• Lin Pan
• Chung-Wei Hang
• Avirup Sil
• Saloni Potdar

We propose a simple and general method to regularize the fine-tuning of Transformer-based encoders for text classification tasks. Specifically, during fine-tuning we generate adversarial examples by perturbing the word embedding matrix of the model and perform contrastive learning on clean and adversarial examples in order to teach the model to learn noiseinvariant representations. By training on both clean and adversarial examples along with the additional contrastive objective, we observe consistent improvement over standard fine-tuning on clean examples. On several GLUE benchmark tasks, our fine-tuned BERTLarge model outperforms BERTLarge baseline by 1. 7% on average, and our fine-tuned RoBERTaLarge improves over RoBERTaLarge baseline by 1. 3%. We additionally validate our method in different domains using three intent classification datasets, where our fine-tuned RoBERTaLarge outperforms RoBERTaLarge baseline by 1–2% on average. For the challenging low-resource scenario, we train our system using half of the training data (per intent) in each of the three intent classification datasets, and achieve similar performance compared to the baseline trained with full training data.

JAAMAS Journal 2011 Journal Article

Generalized framework for personalized recommendations in agent networks

• Chung-Wei Hang
• Munindar P. Singh

Abstract An agent network can be modeled as a directed weighted graph whose vertices represent agents and edges represent a trust relationship between the agents. This article proposes a new recommendation approach, dubbed L oc P at, which can recommend trustworthy agents to a requester in an agent network. We relate the recommendation problem to the graph similarity problem, and define the similarity measurement as a mutually reinforcing relation. We understand an agent as querying an agent network to which it belongs to generate personalized recommendations. We formulate a query into an agent network as a structure graph applied in a personalized manner that reflects the pattern of relationships centered on the requesting agent. We use this pattern as a basis for recommending an agent or object (a vertex in the graph). By calculating the vertex similarity between the agent network and a structure graph, we can produce a recommendation based on similarity scores that reflect both the link structure and the trust values on the edges. Our resulting approach is generic in that it can capture existing network-based approaches merely through the introduction of appropriate structure graphs. We evaluate different structure graphs with respect to two main kinds of settings, namely, social networks and ratings networks. Our experimental results show that our approach provides personalized and flexible recommendations effectively and efficiently based on local information.

TAAS Journal 2011 Journal Article

Trustworthy Service Selection and Composition

• Chung-Wei Hang
• Munindar P. Singh

We consider Service-Oriented Computing (SOC) environments. Such environments are populated with services that stand proxy for a variety of information resources. A fundamental challenge in SOC is to select and compose services, to support specified user needs directly or by providing additional services. Existing approaches for service selection either fail to capture the dynamic relationships between services or assume that the environment is fully observable. In practical situations, however, consumers are often not aware of how the services are implemented. We propose two distributed trust-aware service selection approaches: one based on Bayesian networks and the other on a beta-mixture model. We experimentally validate our approach through a simulation study. Our results show that both approaches accurately punish and reward services in terms of the qualities they offer, and further that the approaches are effective despite incomplete observations regarding the services under consideration.

AAMAS Conference 2009 Conference Paper

Operators for Propagating Trust and their Evaluation in Social Networks

• Chung-Wei Hang
• Yonghong Wang
• Munindar P. Singh

AAMAS Conference 2008 Conference Paper

An Adaptive Probabilistic Trust Model and its Evaluation

• Chung-Wei Hang
• Yonghong Wang
• Munindar Singh

In open settings, the participants are autonomous and there is no central authority to ensure the felicity of their interactions. When agents interact in such settings, each relies upon being able to model the trustworthiness of the agents with whom it interacts. Fundamentally, such models must consider the past behavior of the other parties in order to predict their future behavior. Further, it is sensible for the agents to share information via referrals to trustworthy agents. Much progress has recently been made on probabilistic trust models including those that support the aggregation of information from multiple sources. However, current models do not support trust updates, leaving updates to be handled in an ad hoc manner. This paper proposes a trust representation that combines probabilities and certainty (defined as a function of a probability-certainty density function). Further, it offers a trust update mechanism to estimate the trustworthiness of referrers. This paper describes a testbed that goes beyond existing testbeds to enable the evaluation of a composite probability-certainty model. It then evaluates the proposed trust model showing that the trust model can (a) estimate trustworthiness of damping and capricious agents correctly, (b) update trust values of referrers accurately, and (c) resolve the conflicts in referral networks by certainty discounting.