Min‐Yen Kan lists you as their PhD student. Claim this profile to confirm it and keep the rest of your record right.
Claim this profileAcademic lineage
Traced back 35 generations →Advisors
Works96 from public data
- TAT-QA: A Question Answering Benchmark on a Hybrid of Tabular and Textual Content in Finance652
This work extracts samples from real financial reports to build a new large-scale QA dataset containing both Tabular And Textual data, named TAT-QA, where numerical reasoning is usually required to infer the answer, such as addition, subtraction, multiplication, division, counting, comparison/sorting, and the compositions.
- Advances and challenges in conversational recommender systems: A survey396
A systematic review of the techniques used in current conversational recommender systems, and summarizes the key challenges of developing CRSs in five directions: Question-based user preference elicitation, multi-turn conversational recommendation strategies, dialogue understanding and generation, evaluation and user simulation, and exploitation-exploration trade-offs.
- Estimation-Action-Reflection: Towards Deep Interaction Between Conversational and Recommender Systems353
This work proposes a new CRS framework named Estimation"Action" Reflection, or EAR, which consists of three stages to better converse with users, and conducts extensive experiments on two datasets from Yelp and LastFM to demonstrate significant improvements over the state-of-the-art method CRM.
- Sequicity: Simplifying Task-oriented Dialogue Systems with Single Sequence-to-Sequence Architectures349
A novel, holistic, extendable framework based on a single sequence-to-sequence (seq2seq) model which can be optimized with supervised or reinforcement learning is proposed which significantly outperforms state- of-the-art pipeline-based methods on large datasets and retains a satisfactory entity match rate on out-of-vocabulary (OOV) cases where pipeline-designed competitors totally fail.
- KuaiRec303
A fully-observed dataset from Kuaishou's online environment, where almost all 1,411 users have been exposed to all 3,327 items, and a preliminary analysis of how the two factors - data density and exposure bias - affect the evaluation results of multi-round conversational recommendation.
- KuaiRand: An Unbiased Sequential Recommendation Dataset with Randomly Exposed Videos233
KuaiRand is presented, an unbiased sequential recommendation dataset containing millions of intervened interactions on randomly exposed videos, collected from the video-sharing mobile App, Kuaishou, to enable the research of advanced debiasing large-scale recommendation scenarios for the first time.
- 150
- CIRS: Bursting Filter Bubbles by Counterfactual Interactive Recommender System141
A counterfactual interactive recommender system (CIRS) that augments offline reinforcement learning (offline RL) with causal inference that shows the effectiveness of CIRS in bursting filter bubbles and achieving long-term success in interactive recommendation.
- Seamlessly Unifying Attributes and Items: Conversational Recommendation for Cold-start Users138
The Conversational Thompson Sampling (ConTS) model holistically solves all questions in conversational recommendation by choosing the arm with the maximal reward to play and seamlessly unify attributes and items in the same arm space and achieve their EE trade-offs automatically using the framework ofThompson Sampling.
- Re-examining the Role of Schema Linking in Text-to-SQL130
This work provides a schema linking corpus based on the Spider text-to-SQL dataset, and builds a simple BERT-based baseline to perform a data-driven study on schema linking, finding when schema linking is done well, SLSQL demonstrates good performance on Spider despite its structural simplicity.
- Molweni: A Challenge Multiparty Dialogues-based Machine Reading Comprehension Dataset with Discourse Structure125
The Molweni dataset is presented, a machine reading comprehension (MRC) dataset with discourse structure built over multiparty dialog with large-scale annotations in a modified Segmented Discourse Representation Theory (SDRT; Asher et al., 2016) style.
- Towards Complex Document Understanding By Discrete Reasoning121
A novel model named MHST is developed that takes into account the information in multi-modalities to intelligently address different types of questions with corresponding strategies, i.e., extraction or reasoning.
- Proactive Conversational AI: A Comprehensive Survey of Advancements and Opportunities90
A comprehensive review of various prominent problems and advanced designs for implementing proactivity into different types of dialogue systems, including open-domain dialogues, task-oriented dialogues, and information-seeking dialogues are provided.
- Conversational Recommendation: Formulation, Methods, and Evaluation89
This tutorial covers four emerging directions of conversational recommender systems, providing a review of existing approaches and progress on the topic, and provides take-aways to practitioners to build their own systems.
- 82
- A Unified Multi-task Learning Framework for Multi-goal Conversational Recommender Systems80
This work proposes a novel Unified MultI-goal conversational recommeNDer system (UniMIND), which unify these four tasks with different formulations into the same sequence-to-sequence paradigm and achieves state of the art performance on all tasks with a unified model.
- Towards Analyzing and Understanding the Limitations of DPO: A Theoretical Perspective77
By analyzing the gradient vector field of the DPO loss function, it is found that the DPO loss function decreases the probability of producing human dispreferred data at a faster rate than it increases the probability of producing preferred data.
- Explicit State Tracking with Semi-Supervisionfor Neural Dialogue Generation59
This paper proposes an encoder-decoder architecture, named CopyFlowNet, to represent an explicit dialogue state with a probabilistic distribution over the vocabulary space, and applies a posterior regularization strategy to integrate indirect supervision in the training procedure.
- Tailor: A Soft-Prompt-Based Approach to Attribute-Based Controlled Text Generation57
The proposed Tailor represents each attribute as a pre- trained continuous vector i.e., single-attribute prompt, which guides the generation of a fixed pre-trained language model (PLM) to satisfy aPre-specified attribute.
- SWIM: A Simple Word Interaction Model for Implicit Discourse Relation Recognition54
This work demonstrates that word-weighted averaging can encode argument representation which can be incorporated with word pair information efficiently and saves an order of magnitude in parameters and eschewing the recurrent structure.
- 52
- 50
- Linguistic Properties Matter for Implicit Discourse Relation Recognition: Combining Semantic Interaction, Topic Continuity and Attribution46
This model is built from first principles, analyzing the linguistic properties of the individual top-level Penn Discourse Treebank styled implicit discourse relations: Comparison, Contingency and Expansion to achieve a new state-of-the-art performance level.
- 41
- Rethinking Conversational Agents in the Era of LLMs: Proactivity, Non-collaborativity, and Beyond39
This tutorial will introduce the recent advances on the design of agent’s awareness of goals in a wide range of Conversational systems, including proactive, non-collaborative, and multi-goal conversational systems.
- GraphOTTER: Evolving LLM-based Graph Reasoning for Complex Table Question Answering35
GraphOTTER leverages a graph-based representation, transforming the complex table into an undirected graph, which constructs a clear reasoning path and effectively identifies the answer to a given question.
- Strength Lies in Differences! Improving Strategy Planning for Non-collaborative Dialogues via Diversified User Simulation33
TRIP is proposed to enhance the capability in tailored strategic planning, incorporating a user-aware strategic planning module and a population-based training paradigm and demonstrating the effectiveness of TRIP in catering to diverse users.
- Tailor: A Prompt-Based Approach to Attribute-Based Controlled Text Generation31
The proposed Tailor represents each attribute as a pre-trained continuous vector and guides the generation of a fixed PLM switch to aPre-specified attribute, which can be concatenated with any two single-attribute prompts to multi-attribute text generation.
- Interacting with Non-Cooperative User30
A new solution named I-Pro is proposed that can learn Proactive policy in the Interactive setting via a learned goal weight, which consists of four factors (dialogue turn, goal completion difficulty, user satisfaction estimation, and cooperative degree).
- Dialogue State Tracking with Incremental Reasoning29
This paper proposes to track dialogue states gradually with reasoning over dialogue turns with the help of the back-end data and outperforms the state-of-the-art methods in terms of joint belief accuracy for MultiWOZ 2.1, a large-scale human--human dialogue dataset across multiple domains.
- 26
- Discursive Socratic Questioning: Evaluating the Faithfulness of Language Models’ Understanding of Discourse Relations23
A new method for evaluating the faithfulness of understanding discourse based on question answering, and validates the effec-026 tiveness of explicitly signalled discourse con-027 nectives, the role of contextual information, and the benefits of using historical QA data.
- 23
- GCPG: A General Framework for Controllable Paraphrase Generation23
The combination of lexical and syntactical conditions shows the significant controllable ability of paraphrase generation, and these empirical results could provide novel insight to user-oriented paraphrasing.
- A Dual Prompt Learning Framework for Few-Shot Dialogue State Tracking21
A dual prompt learning framework for few-shot DST is designed that indicates that DST-related knowledge can be probed from pre-trained language models and utilized to address low-resource DST efficiently with the help of prompt learning.
- Goal Awareness for Conversational AI: Proactivity, Non-collaborativity, and Beyond21
This tutorial will introduce the recent advances on the design of agent’s awareness of goals in a wide range of conversational systems and suggest ways to empower the system to handle more complicated conversation tasks that involve strategical and motivational interactions.
- 20
- Concept -- An Evaluation Protocol on Conversational Recommender Systems with System-centric and User-centric Factors19
A new and inclusive evaluation protocol, Concept, which integrates both system- and user-centric factors in CRS, and pinpoints the problem of low usability in the ChatGPT and offers a comprehensive reference guide for evaluating CRS, thereby setting the foundation for CRS improvement.
- Generalized Relation Learning with Semantic Correlation Awareness for Link Prediction19
This work proposes a unified Generalized Relation Learning framework GRL to address the above two problems and enhances the existing link prediction models making them insensitive to unbalanced relation distributions and capable of learning unseen relations.
- GMH: A General Multi-hop Reasoning Model for KG Completion17
This work argues that there are two key issues for a general multi-hop reasoning model: where to go, and when to stop, and proposes a general model which resolves the issues with three modules: the local-global knowledge module to estimate the possible paths, the differentiated action dropout module to explore a diverse set of paths, and the adaptive stopping search module to avoid over searching.
- POS-Constrained Parallel Decoding for Non-autoregressive Generation16
The proposed POS-constrained Parallel Decoding (POSPD) method aims at providing a specific POS sequence to constrain the NAG model during decoding, and demonstrates that POSPD consistently improves NAG models on four text generation tasks to a greater extent compared to knowledge distillation.
- 15
- 14
- Fantastic Expressions and Where to Find Them: Chinese Simile Generation with Multiple Constraints14
This work introduces controllable simile generation (CSG), a new task that requires the model to generate a simile with multiple simile elements, e.g., context and vehicle, and presents GraCe, a model that can serve as a strong baseline for CSG, especially with Scorer, which beats model-based retrieval methods without any re-training.
- Cross-Space Adaptive Filter: Integrating Graph Topology and Node Attributes for Alleviating the Over-smoothing Problem12
This paper proposes a cross-space adaptive filter, called CSF, to produce the adaptive-frequency information extracted from both the topology and attribute spaces, and derives the cross-space filter via an effective multiple-kernel learning strategy, which helps to address the over-smoothing problem while maintaining effectiveness.
- RecSys 2021 Tutorial on Conversational Recommendation: Formulation, Methods, and Evaluation12
This tutorial identifies five emerging trends in the general area of conversational recommender systems: Question-based user preference elicitation; multi-turn conversational recommendation strategies; Dialogue understanding and generation; Exploitation-exploration trade-offs; and Evaluation and user simulation.
- Have We Solved The Hard Problem? It’s Not Easy! Contextual Lexical Contrast as a Means to Probe Neural Coherence12
This paper presents a corpus named Cont 2 Lex to make Contextual Lexical Contrast Recognition a computationally feasible task, and benchmarks this task with widely-adopted semantic representations, finding that contextual embeddings generally outperform static embeds, but barely go beyond 70% in accuracy performance.
- Alibaba-Translate China's Submission for WMT 2022 Metrics Shared Task11
This report builds a system based on the core idea of UNITE (Unified Translation Evaluation), which unifies source- only, reference-only, and source- reference-combined evaluation scenarios into one single model and reduces the gap between pre-training and fine-tuning.
- Constructing Learning Maps for Lecture Videos by Exploring Wikipedia Knowledge11
It is demonstrated that external domain knowledge from Wikipedia can help organize the lecture video corpus and construct more comprehensive knowledge representations, which improves the learning experience of online learners.
- Co-Matching: Towards Human-Machine Collaborative Legal Case Matching9
This work proposes a collaborative matching framework called Co-Matching, which encourages both the machine and the legal practitioner to participate in the matching process, integrating tacit knowledge, and introduces a method called ProtoEM to estimate human decision uncertainty, facilitating the probabilistic combination.
- Effective and Efficient Adversarial Detection for Vision-Language Models via A Single Vector9
A novel and effective iN-time Embedding-based AdveRSarial Image DEtection (NEARSIDE) method, which exploits a single vector that distilled from the hidden states of VLMs, which is called the attacking direction, to achieve the detection of adversarial images against benign ones in the input.
- 9
- Revisit Automatic Error Detection for Wrong and Missing Translation – A Supervised Approach9
A supervised alignment model for translation error detection (AlignDet) is built based on a simple Alignment Triangle strategy to set the benchmark for automatic error detection task and discusses the difficulties and benefits of this task for existing evaluation metrics.
- Abstract Rule Learning for Paraphrase Generation7
In RULER, a novel approach that performs abstract rule learning for paraphrasing by proposing a rule generalizability metric to guide the model to generate rules underlying the paraphrase, and leveraging neural networks to generate paraphrases by refining the sentences transformed by the learned rules.
- 7
- See the Unseen: Better Context-Consistent Knowledge-Editing by Noises6
It is empirically found that the effects of different contexts upon LLMs in recalling the same knowledge follow a Gaussian-like distribution, which can make LLMs see the unseen contexts where the edited knowledge will be applied, therefore improving the editing generalization.
- 6
- Alibaba-Translate China's Submission for WMT 2022 Quality Estimation Shared Task6
This paper’s systems employ the framework of UniTE, which combined three types of input formats during training with a pre-trained language model, to reduce the gap between pre-training and fine-tuning, and use data cropping and a ranking-based score normalization strategy.
- VP-Bench: A Comprehensive Benchmark for Visual Prompting in Multimodal Large Language Models5
VP-Bench is introduced, a benchmark for assessing MLLMs'capability in VP perception and utilization, and establishes a new reference framework for studying how MLLMs comprehend and resolve grounded referring questions.
- 5
- 5
- 5
- 4
- 4
- Dishonesty in Helpful and Harmless Alignment4
It is theoretically shown that the dishonesty can in-turn decrease the alignment performances and augment reward-seeking alignment with representation regularization, and it is theoretically shown that the dishonesty can in-turn decrease the alignment performances and augment reward-seeking alignment with representation regularization.
- DREditor: An Time-efficient Approach for Building a Domain-specific Dense Retrieval Model4
DREditor is developed to edit the matching rule of an off-the-shelf dense retrieval model to suit a specific domain by directly calibrating the output embeddings of the model using an efficient and effective linear mapping.
- Towards Fine-Grained Information: Identifying the Type and Location of Translation Errors4
This paper proposes Fine-Grained Translation Error Detection (FG-TED) task, aiming at identifying both the position and the type of translation errors on given source-hypothesis sentence pairs, and builds an FG-TED model, which can identify both error type and position concurrently.
- OccuQuest: Mitigating Occupational Bias for Inclusive Large Language Models4
An instruction-tuning dataset named OccuQuest is created, which contains 110,000+ prompt-completion pairs and 30,000- dialogues covering over 1,000 occupations in 26 occupational categories and significantly outperforms state-of-the-art LLaMA variants on professional questions in GPT-4 and human evaluations.
- 4
- 3
- AMaPO: Adaptive Margin-attached Preference Optimization for Language Model Alignment2
Adaptive Margin-attached Preference Optimization (AMaPO) employs an instance-wise adaptive margin, refined by Z-normalization and exponential scaling, which dynamically reallocates learning effort by amplifying gradients for misranked samples and suppressing them for correct ones.
- 2
- 2
- 2
- FFHR: Fully and Flexible Hyperbolic Representation for Knowledge Graph Completion2
A Full Poincar\'{e} Multi-relational GCN that achieves graph information propagation in hyperbolic space without requiring any approximation, and ahyperbolic generalization of Euclidean inner product that is beneficial to capture both hierarchical and complex patterns are contributed.
- Knowing-how & Knowing-that: A New Task for Machine Comprehension of User Manuals2
This work introduces the Knowing-how&Knowing-that task that requires the model to answer factoid-style, procedure- style, and inconsistent questions about user manuals, and resolves this task by jointly representing the steps and facts in a graph TARA, which supports a unified inference of various questions.
- RDU: A Region-based Approach to Form-style Document Understanding2
A new KIE model named Region-based Document Understanding (RDU) is developed that takes as input the text content and corresponding coordinates of a document, and tries to predict the result by localizing a bounding-box-like region.
- 1
- 1
- 1
- 1
- 1
- HippoSpark: An On-Demand Experience System for LLM Reasoning–
HippoSpark is introduced, a state-level experience system that performs on-demand retrieval tailored to the immediate needs of the current reasoning state that outperforms both standard prompting and task-level experience baselines.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Vague Preference Policy Learning for Conversational Recommendation–
A novel scenario called Vague Preference Multi-round Conversational Recommendation (VPMCR), which considers users’ vague andvolatile preferences in CRS, and highlights its potential for improving the overall performance andability of CRS in real-world settings, particularly for users with vagueordynamicpreferences.
- DiVa: An Iterative Framework to Harvest More Diverse and Valid Labels from User Comments for Music–
An iterative framework (DiVa) to harvest more diverse and valid labels from user comments for music via pseudo-labels inferred from pre-trained classifiers and a novel joint score function is designed.
- Draft, Command, and Edit: Controllable Text Editing in E-Commerce–
A new draft-command-edit manner in description generation is explored, leading to the proposed new task—controllable text editing in E-commerce and a data augmentation method is designed to remedy the low resource challenge in this task.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.