Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/7194
Title: | 視覺對話:基於影像的中文回覆生成 Visual Chat: Image Grounded Chinese Response Generation |
Authors: | Chi-Wei Chu 朱啟維 |
Advisor: | 傅立成 |
Keyword: | 社交陪伴機器人,聊天系統,聊天機器人,視覺對話, social companion robot,conversation agents,dialog systems,visual chat, |
Publication Year : | 2019 |
Degree: | 碩士 |
Abstract: | 隨著高齡化社會的到來以及勞動力人口缺失的現象,老人陪伴與安養問題漸漸獲得各方的重視。在眾多各種不同的解決方案中,由於其造價相較於人力成本較為低廉,機器人成為了解決這個問題的重要希望。然而,現存的各種陪伴型機器人大多主要提供功能性的幫助,舉凡提醒今日的各項行程,或是報告天氣預報等等。做為一個陪伴型機器人,更甚者,身為家中的一員,機器人的談天能力仍有進步的空間。相比於人們之間的對話,我們能夠自然地將生活周遭事物融入對話中,而機器人卻缺少了這樣的能力。為了賦予機器人這樣的技能,我們利用深度學習的技術來架構一個能夠融合視覺事物的對話系統。除此之外,主動生成回復以及被動生成回復兩個模組,用來處理對話的不同情形。
實驗結果顯示了我們的對話系統具有能夠加入視覺資訊的能力,並且能夠增加對話的豐富度,以吸引人與之對話。 With the gradually aging society and lack of labor, the issue of accompanying elders is getting more and more attention. Robots, due to the low cost as compared to human labor cost, seems to become a promising solution to resolve this issue. Although existing companion robots can provide functional help such as reminding the schedule, reporting the weather forecast, etc. Serving either as a caregiver in home environment or even as a family member, its ability to chat with people still has rooms for improvement. Generally speaking, human-robot conversation is unlike human-human conversations and one of the significant differences is that people are capable of discovering topics from their surroundings, that is, talking about what they see, but robots are still lacking such ability. In order to endow robots with this capability, we develop a dialogue system that can incorporate visual information into dialogue using deep learning techniques. Moreover, the system is composed of active response generation and passive response generation so as to deal with different situations. In order to show that our developed system is efficient, experiments are conducted, which show that our system can indeed incorporate visual information into dialogue, and improve the dialogue content with more information to attract more people to chat with it. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/7194 |
DOI: | 10.6342/NTU201903370 |
Fulltext Rights: | 同意授權(全球公開) |
Appears in Collections: | 電機工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-108-1.pdf | 3.05 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.