東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁 到查詢結果 [ null ]

切換: 標籤 | MARC模式 | ISBD

Unstructured Knowledge Grounding for...

Xu, Yan,

FindBook

Google Book

Amazon

博客來

Unstructured Knowledge Grounding for Open-Domain Dialogues /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Unstructured Knowledge Grounding for Open-Domain Dialogues // Yan Xu.
作者:	Xu, Yan,
面頁冊數:	1 electronic resource (143 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-05, Section: A.
Contained By:	Dissertations Abstracts International86-05A.
標題:	Internet access. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31718355
ISBN:	9798346542957

Unstructured Knowledge Grounding for Open-Domain Dialogues /
Xu, Yan,

Unstructured Knowledge Grounding for Open-Domain Dialogues /Yan Xu. - 1 electronic resource (143 pages)

Source: Dissertations Abstracts International, Volume: 86-05, Section: A.

Open-domain dialogue systems aim at fulfilling human-machine conversations by producing human-like responses to user utterances, offering user experiences across diverse topics. Although data-driven neural methods have excelled in creating intelligent chatbots, the emergence of pretrained language models (PLMs) further turbocharges the development of open-domain dialogue systems. By optimizing PLMs to maximize the token-level likelihood of gold responses given dialogue history, dialogue systems can yield fluent and contextually appropriate responses. Large language model (LLM)-based dialogue systems, such as ChatGPT and GPT-4, demonstrate a superior ability for interaction solely relying on the parametric knowledge from LLMs, which proves the promise of LLM-based chatbots.Despite the strong capabilities of LLM-based dialogue models, challenges still remain: They are not fully reliable. On the one hand, LLM-based dialogue models lack new knowledge. World knowledge is increasing and frequently updated over time. However, LLM-based dialogue models are not exposed to new knowledge. It is also not feasible to continuously post-train the model for new emergent knowledge because of the catastrophic forgetting problem. On the other hand, LLMbased dialogue models suffer from hallucination problems, which produce plausible-sounding but nonsensical content in relation to certain sources. It generates more extrinsic hallucinations from its parametric memory as it does not have access to external knowledge sources.Therefore, we delve into methods for unstructured knowledge grounding in open-domain dialogues to build knowledgeable and trustworthy dialogue systems. Using unstructured knowledge, like natural language corpora, for dialogue response generation poses significant challenges. Specifically, we concentrate on: 1) How to adapt open-domain dialogue models to new topics with unstructured knowledge grounding. While new topics emerge with new knowledge, it is essential for open-domain dialogue models to adapt to new topics. 2) How to craft open-domain dialogue systems with transferable knowledge-grounding skills in niche domains with scarce highquality domain-specific data. As AI applications evolve, there is an emergent demand for building transferable information-seeking dialogue systems in specific domains. Yet, the dearth of highquality domain-specific resources leaves this area largely uncharted and under-explored. 3) How to enhance the faithfulness of responses to external knowledge. Responses from a trustworthy knowledge-grounded dialogue system should be not only informative but also accurate and faithful to external knowledge sources.To elaborate, we firstly propose a groundbreaking model, KnowExpert, that is more knowledgeable about new topics by injecting new knowledge via lightweight adapters and pseudo-conversational style pre-training. The adapters act as knowledge experts for certain topics, allowing the adaptiveness of the model to new topics. We also exploit a topic model for better utilization of the knowledge experts. Secondly, to enable the dialogue model to acquire transferable knowledge-grounding skills, we adopt a modular approach to build a KGD framework for niche domains where the system suffers from limited high-quality domain-specific data resources. The strategy involves decomposing the task into knowledge identification and response generation sub-tasks and employing task-wise pre-training and multi-task learning to reduce the need for high-quality data. Finally, we take a step further to tackle the challenge of response faithfulness to external knowledge. We propose sequential posterior inference (SPI), a novel method that empowers a probabilistic model to effectively learn the inherent correlation between knowledge selection and response generation.

English

ISBN: 9798346542957Subjects--Topical Terms:

3410690
Internet access.

Unstructured Knowledge Grounding for Open-Domain Dialogues /
LDR:05035nmm a22003613i 4500 001 2436762
005 20251103124610.5
006 m o d
007 cr|nu||||||||
008 261011s2023 miu||||||m |||||||eng d
020 $a 9798346542957
035 $a (MiAaPQD)AAI31718355
035 $a AAI31718355
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Xu, Yan, $e author. $3 3827542
245 1 0 $a Unstructured Knowledge Grounding for Open-Domain Dialogues / $c Yan Xu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 1 electronic resource (143 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-05, Section: A.
500 $a Advisors: Fung, Pascale.
502 $b Ph.D. $c Hong Kong University of Science and Technology (Hong Kong) $d 2023.
520 $a Open-domain dialogue systems aim at fulfilling human-machine conversations by producing human-like responses to user utterances, offering user experiences across diverse topics. Although data-driven neural methods have excelled in creating intelligent chatbots, the emergence of pretrained language models (PLMs) further turbocharges the development of open-domain dialogue systems. By optimizing PLMs to maximize the token-level likelihood of gold responses given dialogue history, dialogue systems can yield fluent and contextually appropriate responses. Large language model (LLM)-based dialogue systems, such as ChatGPT and GPT-4, demonstrate a superior ability for interaction solely relying on the parametric knowledge from LLMs, which proves the promise of LLM-based chatbots.Despite the strong capabilities of LLM-based dialogue models, challenges still remain: They are not fully reliable. On the one hand, LLM-based dialogue models lack new knowledge. World knowledge is increasing and frequently updated over time. However, LLM-based dialogue models are not exposed to new knowledge. It is also not feasible to continuously post-train the model for new emergent knowledge because of the catastrophic forgetting problem. On the other hand, LLMbased dialogue models suffer from hallucination problems, which produce plausible-sounding but nonsensical content in relation to certain sources. It generates more extrinsic hallucinations from its parametric memory as it does not have access to external knowledge sources.Therefore, we delve into methods for unstructured knowledge grounding in open-domain dialogues to build knowledgeable and trustworthy dialogue systems. Using unstructured knowledge, like natural language corpora, for dialogue response generation poses significant challenges. Specifically, we concentrate on: 1) How to adapt open-domain dialogue models to new topics with unstructured knowledge grounding. While new topics emerge with new knowledge, it is essential for open-domain dialogue models to adapt to new topics. 2) How to craft open-domain dialogue systems with transferable knowledge-grounding skills in niche domains with scarce highquality domain-specific data. As AI applications evolve, there is an emergent demand for building transferable information-seeking dialogue systems in specific domains. Yet, the dearth of highquality domain-specific resources leaves this area largely uncharted and under-explored. 3) How to enhance the faithfulness of responses to external knowledge. Responses from a trustworthy knowledge-grounded dialogue system should be not only informative but also accurate and faithful to external knowledge sources.To elaborate, we firstly propose a groundbreaking model, KnowExpert, that is more knowledgeable about new topics by injecting new knowledge via lightweight adapters and pseudo-conversational style pre-training. The adapters act as knowledge experts for certain topics, allowing the adaptiveness of the model to new topics. We also exploit a topic model for better utilization of the knowledge experts. Secondly, to enable the dialogue model to acquire transferable knowledge-grounding skills, we adopt a modular approach to build a KGD framework for niche domains where the system suffers from limited high-quality domain-specific data resources. The strategy involves decomposing the task into knowledge identification and response generation sub-tasks and employing task-wise pre-training and multi-task learning to reduce the need for high-quality data. Finally, we take a step further to tackle the challenge of response faithfulness to external knowledge. We propose sequential posterior inference (SPI), a novel method that empowers a probabilistic model to effectively learn the inherent correlation between knowledge selection and response generation.
546 $a English
590 $a School code: 1223
650 4 $a Internet access. $3 3410690
650 4 $a Large language models. $3 3775019
650 4 $a Web studies. $3 2122754
690 $a 0800
690 $a 0646
710 2 $a Hong Kong University of Science and Technology (Hong Kong). $e degree granting institution. $3 3814653
720 1 $a Fung, Pascale $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-05A.
790 $a 1223
791 $a Ph.D.
792 $a 2023
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31718355