Existing solutions to task-oriented dialogue systems follow pipeline designs which introduces architectural complexity and fragility. We propose a novel, holistic, extendable framework based on single sequence-to-sequence (seq2seq) model can be optimized with supervised or reinforcement learning. A key contribution is that we design text spans named belief track believes, allowing modeled in seq2seq way. Based this, simplistic Two Stage CopyNet instantiation emonstrates good scalability:...

Recommender systems exploit interaction history to estimate user preference, having been heavily used in a wide range of industry applications. However, static recommendation models are difficult answer two important questions well due inherent shortcomings: (a) What exactly does like? (b) Why like an item? The shortcomings the way that learn i.e., without explicit instructions and active feedback from users. recent rise conversational recommender (CRSs) changes this situation fundamentally....

Traditional recommendation systems estimate user preference on items from past interaction history, thus suffering the limitations of obtaining fine-grained and dynamic preference. Conversational system (CRS) brings revolutions to those by enabling directly ask users about their preferred attributes items. However, existing CRS methods do not make full use such advantage -- they only attribute feedback in rather implicit ways as updating latent representation. In this paper, we propose Path...

Open-domain Question Answering (OpenQA) is an important task in Natural Language Processing (NLP), which aims to answer a question the form of natural language based on large-scale unstructured documents. Recently, there has been surge amount research literature OpenQA, particularly techniques that integrate with neural Machine Reading Comprehension (MRC). While these works have advanced performance new heights benchmark datasets, they rarely covered existing surveys QA systems. In this...

While personalization increases the utility of recommender systems, it also brings issue filter bubbles . e.g., if system keeps exposing and recommending items that user is interested in, may make feel bored less satisfied. Existing work studies in static recommendation, where effect overexposure hard to capture. In contrast, we believe more meaningful study interactive recommendation optimize long-term satisfaction. Nevertheless, unrealistic train model online due high cost. As such, have...

Recommender systems are embracing conversational technologies to obtain user preferences dynamically, and overcome inherent limitations of their static models. A successful Conversational System (CRS) requires proper handling interactions between conversation recommendation. We argue that three fundamental problems need be solved: 1) what questions ask regarding item attributes, 2) when recommend items, 3) how adapt the users' online feedback. To best our knowledge, there lacks a unified...

Fengbin Zhu, Wenqiang Lei, Youcheng Huang, Chao Wang, Shuo Zhang, Jiancheng Lv, Fuli Feng, Tat-Seng Chua. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Emerging research in Neural Question Generation (NQG) has started to integrate a larger variety of inputs, and generating questions requiring higher levels cognition. These trends point NQG as bellwether for NLP, about how human intelligence embodies the skills curiosity integration. We present comprehensive survey neural question generation, examining corpora, methodologies, evaluation methods. From this, we elaborate on what see emerging NQG's trend: terms learning paradigms, input...

Static recommendation methods like collaborative filtering suffer from the inherent limitation of performing real-time personalization for cold-start users. Online recommendation, e.g., multi-armed bandit approach, addresses this by interactively exploring user preference online and pursuing exploration-exploitation (EE) trade-off. However, existing bandit-based model actions homogeneously. Specifically, they only consider items as arms, being incapable handling item attributes, which...

The progress of recommender systems is hampered mainly by evaluation as it requires real-time interactions between humans and systems, which too laborious expensive. This issue usually approached utilizing the interaction history to conduct offline evaluation. However, existing datasets user-item are partially observed, leaving unclear how what extent missing will influence To answer this question, we collect a fully-observed dataset from Kuaishou's online environment, where almost all 1,411...

Recommender systems deployed in real-world applications can have inherent exposure bias, which leads to the biased logged data plaguing researchers. A fundamental way address this thorny problem is collect users' interactions on randomly expose items, i.e., missing-at-random data. few works asked certain users rate or select recommended e.g., Yahoo!, Coat, and OpenBandit. However, these datasets are either too small size lack key information, such as unique user ID features of users/items....

Recent years witnessed several advances in developing multi-goal conversational recommender systems (MG-CRS) that can proactively attract users’ interests and naturally lead user-engaged dialogues with multiple goals diverse topics. Four tasks are often involved MG-CRS, including Goal Planning, Topic Prediction, Item Recommendation, Response Generation. Most existing studies address only some of these tasks. To handle the whole problem modularized frameworks adopted where each task is...

Research into the area of multiparty dialog has grown considerably over recent years. We present Molweni dataset, a machine reading comprehension (MRC) dataset with discourse structure built dialog. Molweni’s source samples from Ubuntu Chat Corpus, including 10,000 dialogs comprising 88,303 utterances. annotate 30,066 questions on this corpus, both answerable and unanswerable questions. also uniquely contributes dependency annotations in modified Segmented Discourse Representation Theory...

Recommender systems have demonstrated great success in information seeking. However, traditional recommender work a static way, estimating user preferences on items from past interaction history. This prevents capturing dynamic and fine-grained of users. Conversational bring revolution to existing systems. They are able communicate with users through natural languages during which they can explicitly ask whether likes an attribute or not. With the preferred attributes, system conduct more...

In existing sophisticated text-to-SQL models, schema linking is often considered as a simple, minor component, belying its importance. By providing corpus based on the Spider dataset, we systematically study role of linking. We also build simple BERT-based baseline, called Schema-Linking SQL (SLSQL) to perform data-driven study. find when done well, SLSQL demonstrates good performance despite structural simplicity. Many remaining errors are attributable noise. This suggests crux for current...

Knowledge graph completion (KGC) has become a focus of attention across deep learning community owing to its excellent contribution numerous downstream tasks. Although recently have witnessed surge work on KGC, they are still insufficient accurately capture complex relations, since adopt the single and static representations. In this work, we propose novel Disentangled Graph Attention Network (DisenKGAT) for which leverages both micro-disentanglement macro-disentanglement exploit...

Conversational recommendation system (CRS) attracts increasing attention in various application domains such as retail and travel. It offers an effective way to capture users' dynamic preferences with multi-turn conversations. However, most current studies center on the aspect while over-simplifying conversation process. The negligence of complexity data structure flow hinders their practicality utility. In reality, there exist relationships among slots values, requirements may dynamically...

Conversational systems based on Large Language Models (LLMs), such as ChatGPT, show exceptional proficiency in context understanding and response generation. However, they still possess limitations, failing to ask clarifying questions ambiguous queries or refuse users' unreasonable requests, both of which are considered key aspects a conversational agent's proactivity. This raises the question whether LLM-based equipped handle proactive dialogue problems. In this work, we conduct...

Proactive dialogue systems, related to a wide range of real-world conversational applications, equip the agent with capability leading conversation direction towards achieving pre-defined targets or fulfilling certain goals from system side. It is empowered by advanced techniques progress more complicated tasks that require strategical and motivational interactions. In this survey, we provide comprehensive overview prominent problems designs for agent's proactivity in different types...

The task of dialogue generation aims to automatically provide responses given previous utterances. Tracking states is an important ingredient in for estimating users' intention. However, the expensive nature state labeling and weak interpretability make tracking a challenging problem both task-oriented non-task-oriented generation: For generating dialogues, usually learned from manually annotated corpora, where human annotation training; most existing work neglects explicit due unlimited...

Capturing the semantic interaction of pairs words across arguments and proper argument representation are both crucial issues in implicit discourse relation recognition. The current state-of-the-art represents as distributional vectors that computed via bi-directional Long Short-Term Memory networks (BiLSTMs), known to have significant model complexity.In contrast, we demonstrate word-weighted averaging can encode which incorporate word pair information efficiently. By saving an order...

A watt-level traveling wave tube (TWT) amplifier operating above 0.3 THz has been demonstrated. This TWT based on the folded waveguide (FWG) slow-wave structure achieved over 3.1 W of output power at 336.96 GHz, and corresponding device gain achieves 26.2 dB. In this letter, design experimental test results are presented.