發(fā)布時間：2018-11-22 11:51

【摘要】：身處互聯(lián)網(wǎng)飛速發(fā)展的時代,京東、天貓和亞馬遜等在線購物網(wǎng)站在人們的生活中扮演著越來越重要的角色,網(wǎng)上購物成為了重要的購買方式。在網(wǎng)上購物時人們往往通過三個途徑獲取商品信息,圖片、產(chǎn)品參數(shù)和評論。賣家已經(jīng)美化過圖片中隱藏的商品信息,產(chǎn)品參數(shù)可能過于專業(yè)化,并非所有人都可以看懂,評論數(shù)據(jù)的可讀性與豐富性使得評論往往會成為顧客決定是否購買的標尺。但是評論數(shù)量是巨大的,如何將這些評論有效整理并建立商品評價模型,幫助顧客挑選商品、幫助賣家改進產(chǎn)品是本文研究的重點。以往的商品評價模型主要有兩類,一類是基于產(chǎn)品參數(shù),該方法認為產(chǎn)品的好壞完全是由硬件決定的,忽視了顧客的使用體驗,當然省時省力是該方法的優(yōu)點。另一類是基于問卷調(diào)查,該方法將顧客的感覺放在了第一位,但是問卷的設(shè)計、發(fā)放、回收和整理的過程耗時耗力。而筆者建立的基于評論數(shù)據(jù)的商品評價模型有著省時省力和貼合用戶使用體驗的優(yōu)點。本文在建立商品評價模型時主要完成以下工作:1.數(shù)據(jù)的獲取與清洗。利用python對電商網(wǎng)站的評論數(shù)據(jù)進行爬取,定制相應(yīng)爬蟲規(guī)則。重復(fù)的獲取數(shù)據(jù)、虛假評論的重復(fù)性和無意義評論之間的相似性,為了減少以上三種情況對于最終評價模型的影響,筆者這利用文本相似度計算對評論數(shù)據(jù)進行了清洗。2.情感單元的抽取。本文使用基于詞典匹配的情感單元提取模型,將不規(guī)則的評論數(shù)據(jù)轉(zhuǎn)化成規(guī)范的問卷式數(shù)據(jù)。為了提高情感抽取的準確性和完整性,筆者使用Apriori模型擴充知網(wǎng)提供的正負面評價詞典,最終評估發(fā)現(xiàn)該情感模型對于短句情感單元抽取的正確率已經(jīng)達到90%。3.商品評價模型的建立。即利用LDA模型對評論進行分析,找出評論中潛在主題建立指標體系。接著為了使高質(zhì)量高認可的評論對于商品最終評價結(jié)果影響更大,建立了評價的有效度模型,最終選用了模糊評價模型對商品進行評價分析,模糊矩陣的構(gòu)造則依靠有效度模型的結(jié)果。筆者使用三部小米手機的評論建立基于商品評論的評價模型,通過評價結(jié)果可以知道電池容量和手機屏幕方面小米max略勝一籌,與產(chǎn)品參數(shù)非常一致。在照相功能上,單純考慮手機參數(shù)小米5s應(yīng)該獲得第一,但是評價結(jié)果卻是小米5s惜敗于小米5,通過分析評論發(fā)現(xiàn)小米5s拍照會出現(xiàn)無法對焦、輕微抖動照片不清晰和像素不夠的問題。通過分析評價結(jié)果可以發(fā)現(xiàn),筆者結(jié)合爬蟲、情感分析技術(shù)和統(tǒng)計知識建立的基于情感分析的商品評價模型,既省時省力,評價結(jié)果也非常貼合顧客使用體驗。
[Abstract]:In the era of the rapid development of the Internet, online shopping sites such as JingDong, Tmall and Amazon are playing an increasingly important role in people's lives, and online shopping has become an important way to buy. When shopping online, people often get product information, pictures, product parameters and comments through three ways. The seller has beautified the hidden product information in the picture, the product parameter may be too specialized, not everyone can understand. The readability and richness of the comment data make the comment often become the yardstick that the customer decides whether to buy or not. However, the number of comments is huge. How to organize these comments effectively and establish a commodity evaluation model to help customers select products and help sellers to improve their products is the focus of this paper. There are two main types of commodity evaluation models in the past. One is based on product parameters. This method holds that the quality of product is completely determined by hardware and neglects the experience of customers. Of course, saving time and effort is the advantage of this method. The other is based on questionnaire, which puts the customer's feeling first, but the design, distribution, recovery and finishing of the questionnaire are time-consuming and laborious. The commodity evaluation model based on comment data has the advantages of saving time and labor and fitting the user's experience. The main work of this paper is as follows: 1. Data acquisition and cleaning. Using python to crawl the comment data of ecommerce website and customize the corresponding crawler rules. In order to reduce the influence of the above three cases on the final evaluation model, the author uses the text similarity calculation to clean the comment data. The extraction of emotional units. In this paper, an emotional unit extraction model based on dictionary matching is used to transform irregular comment data into standardized questionnaire data. In order to improve the accuracy and completeness of emotion extraction, the author uses Apriori model to expand the dictionary of positive and negative evaluation provided by Zhiwang. Finally, it is found that the correct rate of emotion model for extracting short sentence emotional units has reached 90. 3. The establishment of commodity evaluation model. That is to use LDA model to analyze comments and find out the potential topics in the comments to establish an index system. Then, in order to make the high quality and high recognition comments have more influence on the final evaluation results, the validity model of evaluation is established, and the fuzzy evaluation model is used to evaluate and analyze the goods. The construction of fuzzy matrix depends on the result of effectiveness model. The evaluation model based on commodity review is established by using the comments of three Xiaomi mobile phones. Through the evaluation results, we can know that Xiaomi max is superior in battery capacity and mobile phone screen, which is very consistent with the product parameters. In terms of photographic function, only considering the mobile phone parameter Xiaomi 5s should get the first place, but the evaluation result is that Xiaomi 5s loses to Xiaomi 5s. Through analysis and comments, it is found that Xiaomi 5s will not be able to focus when taking pictures. Slightly jitter the picture is not clear and the pixel is not enough problems. Through the analysis of the evaluation results, we can find that the commodity evaluation model based on emotion analysis, which is based on crawler, emotion analysis technology and statistical knowledge, not only saves time and effort, but also fits the customer experience very well.
【學(xué)位授予單位】：安徽財經(jīng)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：F713.36

【參考文獻】

相關(guān)期刊論文 前10條

1 張克亮;黃金柱;曹蓉;李峰;;基于HNC語境框架和情感詞典的文本情感傾向分析[J];山東大學(xué)學(xué)報(理學(xué)版);2016年07期

2 聶卉;吳毅駿;;基于特征表現(xiàn)的虛假評論人預(yù)測研究[J];圖書情報工作;2015年10期

3 王博;劉盛博;丁X;劉則淵;;基于LDA主題模型的專利內(nèi)容分析方法[J];科研管理;2015年03期

4 周練;;Word2vec的工作原理及應(yīng)用探究[J];科技情報開發(fā)與經(jīng)濟;2015年02期

5 蔣翠清;梁坤;丁勇;劉士喜;劉堯;;基于社會媒體的股票行為預(yù)測[J];中國管理科學(xué);2015年01期

6 陳磊磊;;不同距離測度的K-Means文本聚類研究[J];軟件;2015年01期

7 陳燕方;李志宇;;基于評論產(chǎn)品屬性情感傾向評估的虛假評論識別研究[J];現(xiàn)代圖書情報技術(shù);2014年09期

8 錢智勇;周建忠;童國平;蘇新寧;;基于HMM的楚辭自動分詞標注研究[J];圖書情報工作;2014年04期

9 付沙;周航軍;;關(guān)聯(lián)規(guī)則挖掘Apriori算法的研究與改進[J];微電子學(xué)與計算機;2013年09期

10 李志宇;;在線商品評論效用排序模型研究[J];現(xiàn)代圖書情報技術(shù);2013年04期

相關(guān)碩士學(xué)位論文 前2條

1 金麗君;基于SVM的搜索型商品評論有用性自動識別方法研究[D];哈爾濱工業(yè)大學(xué);2013年

2 李明;針對特定領(lǐng)域的中文新詞發(fā)現(xiàn)技術(shù)研究[D];南京航空航天大學(xué);2012年

，

本文編號：2349297