【摘要】：隨著網(wǎng)絡(luò)和計(jì)算機(jī)在人們生活中的普及,消費(fèi)者在網(wǎng)上購物后會(huì)通過網(wǎng)絡(luò)發(fā)表自己對(duì)產(chǎn)品的評(píng)論。由于這些評(píng)論包含了消費(fèi)者對(duì)產(chǎn)品的評(píng)價(jià)意見,因此對(duì)其進(jìn)行挖掘?qū)ι碳腋倪M(jìn)產(chǎn)品以及消費(fèi)者購買產(chǎn)品起到了輔助作用,有著廣泛的應(yīng)用價(jià)值和研究?jī)r(jià)值。 產(chǎn)品評(píng)論挖掘主要包括產(chǎn)品特征抽取、觀點(diǎn)抽取以及極性分類等方面的研究。而在這些研究中,以往的方法將研究重點(diǎn)集中在評(píng)論內(nèi)容上,同時(shí)所依賴的極性詞典也是由固定詞構(gòu)成的,因而影響了評(píng)論挖掘的效果。針對(duì)這些問題,本文面向圖書評(píng)論進(jìn)行了研究,主要工作包括以下幾個(gè)方面: 在構(gòu)建詞典時(shí),考慮到詞對(duì)類別的貢獻(xiàn)不同,將CHI值思想應(yīng)用到詞典構(gòu)建中,提出了一種基于改進(jìn)CHI值的極性詞典構(gòu)建方法。該方法通過計(jì)算每個(gè)詞的CHI值來完成對(duì)詞的極性分類。隨后,對(duì)于沒有包含在詞典中的詞,根據(jù)同類極性詞共現(xiàn)的特點(diǎn)對(duì)其進(jìn)行提取并將滿足閾值要求的添加到相應(yīng)的極性詞典中,實(shí)現(xiàn)了詞典的動(dòng)態(tài)添加,在一定程度上解決了詞典固定不變的問題。另外,考慮到有些極性詞是特定修飾某一特征的,因此又將極性詞做了進(jìn)一步地劃分,以便用于分析未包含特征的評(píng)論。 在進(jìn)行評(píng)論極性分析時(shí),改進(jìn)了轉(zhuǎn)折復(fù)句的極性計(jì)算公式,以適用于圖書評(píng)論。同時(shí),考慮到某些圖書評(píng)論帶有標(biāo)題,而這些標(biāo)題通常表達(dá)了評(píng)論者的觀點(diǎn)傾向,據(jù)此提出了基于標(biāo)題和改進(jìn)的重轉(zhuǎn)句極性計(jì)算公式的評(píng)論極性分析方法。該方法將標(biāo)題極性作為評(píng)論的極性標(biāo)注來對(duì)其進(jìn)行極性分析,并在分析中利用上述改進(jìn)的公式調(diào)整評(píng)論極性,從而減少了評(píng)論極性分析的錯(cuò)誤。 在進(jìn)行評(píng)論觀點(diǎn)總結(jié)時(shí),改進(jìn)了SBV算法,以適用于圖書評(píng)論。該方法主要根據(jù)詞語間的依存關(guān)系來提取評(píng)論句中的特征和觀點(diǎn),據(jù)此對(duì)評(píng)論進(jìn)行觀點(diǎn)總結(jié)。 實(shí)驗(yàn)結(jié)果表明,本文提出的方法是有效的,較好地改善了圖書評(píng)論觀點(diǎn)分析的效果。
[Abstract]:With the popularity of the Internet and computers in people's lives, consumers will post their comments on products after shopping online. Because these comments contain consumers' opinions on product evaluation, mining them plays an auxiliary role in improving products and consumers' purchasing products, and has extensive application value and research value. Product comment mining includes product feature extraction, viewpoint extraction and polarity classification. In these studies, the previous methods focus on the content of comments, and the polarity dictionary which depends on is also composed of fixed words, which affects the effect of comment mining. In order to solve these problems, this paper focuses on the book review. The main work includes the following aspects: in the construction of dictionaries, considering the different contribution of words to categories, the idea of CHI value is applied to the construction of dictionaries. A polarity dictionary construction method based on improved CHI value is proposed. The polarity classification of each word is accomplished by calculating the CHI value of each word. Then, for the words not included in the dictionary, according to the co-occurrence characteristics of the same polar words, we extract them and add them to the corresponding polar dictionaries that meet the threshold requirements, so that the dynamic addition of the dictionaries is realized. To some extent, the problem of fixed dictionary is solved. In addition, considering that some polarity words are specific to a certain feature, polarity words are further divided to analyze comments that do not contain features. In the analysis of the polarity of comments, the formula for calculating polarity of transition complex sentences is improved to apply to book reviews. At the same time, considering that some book reviews have titles, which usually express the opinion tendency of the reviewers, a method for analyzing the polarity of comments based on the title and improved formula for calculating the polarity of repetition sentences is proposed. In this method, the polarity of a comment is labeled as the polarity of a comment, and the polarity of the comment is adjusted by using the improved formula in the analysis, thus reducing the errors in the analysis of the polarity of the comment. In the summary of comments, the SBV algorithm is improved to be suitable for book review. This method is mainly based on the dependency relationship between words to extract the features and viewpoints of comment sentences, and then summarize the opinions of comments. The experimental results show that the method proposed in this paper is effective and can improve the analysis effect of book review.
【學(xué)位授予單位】：河北大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2011
【分類號(hào)】：G236

【引證文獻(xiàn)】

相關(guān)碩士學(xué)位論文 前1條

1 付瓊芳;基于網(wǎng)上產(chǎn)品評(píng)論挖掘系統(tǒng)設(shè)計(jì)與實(shí)現(xiàn)[D];暨南大學(xué);2012年

，

本文編號(hào)：2273062