Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69745
Title: | 評價語言分析與主題擷取-網路旅遊部落格情感分析之應用 Evaluative Language and Topic Extraction: An Aspect-based Sentiment Analysis on Online Travel Blogs |
Authors: | Chia-Chen Lee 李佳臻 |
Advisor: | 謝舒凱(Shu-Kai Hsieh) |
Keyword: | 情感分析,評價語言,主題模型,模式文法,意見偵測,意見擷取, aspect-based sentiment analysis,evaluative language,opinion extraction,opinion mining,pattern grammar,topic modeling, |
Publication Year : | 2018 |
Degree: | 碩士 |
Abstract: | 本論文研究旨在探究網路旅遊部落格之評價語言,分析其文章評價語言及情感分析,並以語言學分析為本之特徵輔以主題模型(Topic modeling)機器學習,進行以篇章為單位之評價表達內容及被評價目標之擷取(aspect-based sentiment analysis)。觀察透過語言學理論分析出之特徵,搭配機器學習法是否有助於評價之擷取。旅遊部落客表達之意見、看法及評價為讀者規劃旅遊時幫助決策進行的重要參考指標。本文利用模式文法(Pattern Grammar)歸納出特定情感極度的句構組合,只要在此文本領域出現皆呈現為正或負向極度,作為推薦與否的表達。相關該領域之特定評價用詞及評價促發詞同樣為重要之特徵。透過主題模型訓練分析篇章中可能涵蓋之主題(被評價目標類型),以及定義之被評價目標層次(aspect),擷取出分類更細緻之被評價目標面向及評價內容。本文期望作為一個評價語言分析及擷取應用之架構,適用於文本類型相似的其他領域,透過模式文法輔以主題模型訓練之混合法,擷取文章內之評價精華,並能作為意見偵測及文文章摘要等相關重要應用之基礎。 This study aims to explore evaluative language and sentiment analysis, based on online travel blogs. Aspect-based sentiment analysis is employed for document-based opinion target expression (aspect) and opinion expression extraction. Linguistic features are analyzed for extraction task, with the assist of Topic Modeling technique to examine the applicability of such hybrid method. Opinions, comments, and evaluative expressions in travel blog articles are primary sources for people to make travel decision. This study employs Pattern Grammar for analyzing domain-restricted evaluative patterns with fixed sentiment polarity. Aspect-restricted opinion expressions and opinion triggering words are also used for extraction. Latent Dirichlet Allocation is used for extracting potential topics in a given document, to assist in extracting finer-grained aspects and its corresponding opinion expressions. This study expects to serve as a framework for similar type of contents in other domains, with linguistic-based analysis such as pattern grammar, and topic modeling method to better extract opinion target expressions and opinion expressions. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/69745 |
DOI: | 10.6342/NTU201800804 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 語言學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-107-1.pdf Restricted Access | 2.88 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.