Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (3 results)

  • Thesis
  • Peer Reviewed
UC Santa Barbara Electronic Theses and Dissertations (2022)

As the primary means of human communication, natural language bears the functionality to bridge the abstract knowledge form and its realizations to all kinds of human usages. In an artificial intelligence assistant, the natural language interface is grounded on various knowledge sources and designated to perform conveyance, navigation, and reasoning among those sources. Based on the mechanism to interact with the knowledge groundings, the natural language processing tasks can be primarily categorized into 1) natural language understanding (NLU): understanding the semantics of the natural language from the user to perform navigation or reasoning among the knowledge sources, and 2) natural language generation (NLG): generating natural language responses to the user grounded on the knowledge sources. In this thesis, I will elaborate on my work on both NLU and NLG divisions in order to extend the edge of current research toward building an advanced, human-like AI assistant.

For the NLU division, first, current large pre-trained language models can achieve very strong performances on many NLP tasks but still struggle with tasks requiring complex reasoning abilities. In this thesis, I will discuss my work on complex numerical reasoning over structured and unstructured knowledge. Second, current dialogue assistants follow a fixed ontology to represent user preferences, which makes them essentially agent-centric and lack the power to represent fine-grained, realistic user preferences. In this thesis, I will discuss our first step to build the user-centric dialogue system by proposing a new formulation for user preference representations.

For the NLG division, first, current models can obtain strong performances with large-scale training data, but this is not realistic for real-world applications. In this thesis, I will discuss my work on few-shot natural language generation from tabular data. Second, NLG models often suffer from factual correctness when generating descriptions with logical inferences. This thesis proposes to study high-fidelity natural language generation with logical inferences. At last, I will discuss my work for NLG in the important real application domain of dialogue systems. Current research on dialogue systems typically studies each type of dialogue individually. However, the ultimate goal of conversational AI is to build a unified assistant capable of conversing with all kinds of dialogues and switching among them seamlessly. Towards this goal, this thesis studies to enrich task-oriented dialogue with knowledge-grounded chitchat.

In the above work, we define the novel tasks and propose new datasets, as well as design approaches to tackle the new challenges. Finally, we will make the conclusion and discuss the future plans and directions.

    Cover page: Knowledge-Grounded Natural Language Processing
    • Article
    • Peer Reviewed
    UCLA Previously Published Works (2015)
    Contact-angle hysteresis of a liquid suspended on surface microstructures, namely in a Cassie-Baxter state, is determined mainly by the receding contact line although not fully understood. Existing modified Cassie-Baxter models predict some but not most experimental data in the literature. Noting that most models were based on the two-dimensional (2-D) principle whereas the experiments were under three-dimensional (3-D) conditions, here we develop a 2-D experiment. While 3-D experiments measure the receding contact lines averaged over space and time, 2-D experiments eliminate the spatial averaging and can further eliminate the temporal averaging by high-speed visualization. The resulting details of the contact line motion lead us to propose a 2-D model, which incorporates the contact-line friction. The new 2-D model matches the 2-D experimental results excellently while all existing models show significant deviation. By introducing a line solid fraction term, the 2-D model is further generalized to a 3-D model, which successfully predicts a wide range of 3-D data in the literature regardless of their distinct microstructures and receding modes.
    • 1 supplemental PDF
    • 3 supplemental videos
    Cover page: A dynamic Cassie–Baxter model
    • Article
    • Peer Reviewed
    Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 46 (2024)
    Next point-of-interest (POI) recommendation is a challenging task that aims to recommend the next location that a user may be interested in based on their check-in trajectories. Since users travel not only with long-term stable preferences but also with short-term dynamic interests, there is often a potential dependency between long-term and short-term preferences. Most existing works tend to mine the dependencies between long-term and short-term trajectories by contrastive learning but always ignore the negative impact of the learned dependencies on the accuracy of short-term trajectory modeling. Moreover, they often only utilize the context information of the user's trajectory, while neglecting the spatiotemporal dependencies between user trajectories. To address these issues, we proposed a novel dual contrastive learning framework DCLS. Specifically, we designed a novel dual contrastive learning scheme, for which we built two views: the first view is between the user's own long-term and short-term trajectories, and the second view is between the short-term trajectories of different users. We performed contrastive learning on both views, to learn the dependency between long-term and short-term trajectories, and improve the accuracy of trajectory modeling. We also designed a multi-class attention fusion module, which integrates the spatiotemporal influence of trajectory dependencies on user mobility, enhancing the recommendation performance. We conducted extensive experiments on three real-world datasets, which demonstrated that our model achieves advanced performance in the next POI recommendation.
      Cover page: Dual Contrastive Learning for Next POI Recommendation with Long and Short-Term Trajectory ModelingCreative Commons 'BY' version 4.0 license
      Top