社會媒體中的用戶偏好建模研究
發(fā)布時間：2018-08-01 14:44
【摘要】：隨著Web 2.0的發(fā)展,人們已經習慣在網(wǎng)上發(fā)表自己的觀點及看法,也從別人發(fā)布的信息中獲取自己所需的信息,從而形成了一個由廣大用戶主導的互聯(lián)網(wǎng)模式。在這樣的互聯(lián)網(wǎng)模式下,人們越來越依賴網(wǎng)絡,從最初的查找資料,到后來的各種聊天社區(qū),到現(xiàn)在衣食住行等都要到網(wǎng)上看別人的評價才會做出決定,互聯(lián)網(wǎng)正在改變人們生活的方方面面。而社會媒體正是這些行為的媒介,包括虛擬社區(qū)和網(wǎng)絡平臺等,人們可以在上面創(chuàng)作、分享、交流意見、觀點及經驗,主要包括微博、博客、論壇、網(wǎng)絡社區(qū)、評論網(wǎng)站等。人們在社會媒體上發(fā)表自己的觀點,而個人觀點一般是帶有情感偏好的,這些觀點大致分為2類,一類是文本信息,比如微博的內容等,另一類是打分信息,比如電影的評分等。用戶偏好是指用戶對于某件事件、物品的喜愛、厭惡等各種情感。用戶偏好研究就是通過研究這些蘊含了豐富情感的信息,了解用戶想表達的情感偏好。本文將從方面評分的評分預測和唐代詩詞的情感分析兩個方面來研究社會媒體中的用戶偏好問題。方面評分是產品各個細致方面的評分,而總評分是產品所有方面的綜合評分�，F(xiàn)今,大部分使用總評分的工作都是基于這樣一個假設:總評分是方面評分的平均分或總評分與方面評分很接近。然而經過分析真實數(shù)據(jù)集發(fā)現(xiàn),在總評分和方面評分之間存在一個評分偏差,但現(xiàn)有工作并沒有考慮評分偏差。本文首次研究了帶有評分偏差的方面評分預測問題,提出了一個新的情感主題混合模型RCMB。RCMB認為總評分是概率圖的中心并通過一個隱藏方面評分變量整合了評分偏差先驗信息。在真實數(shù)據(jù)集(大眾點評和TripAdvisor)上的實驗表明,RCMB比其他現(xiàn)有方法取得了更高的預測準確率,并更能保持評分的相對順序�，F(xiàn)有的情感分析工作一般是關注現(xiàn)代文本,比如產品評論和微博,很少涉及古代文學作品的分析。而詩詞則相當于古代人所使用的微博,也是表達其情感的重要媒介。本文提出了一個基于遷移學習的中國唐代詩歌情感分類模型TL-PCO,通過分析詩歌的情感可以了解到當時的社會和文化進展。TL-PCO通過兩個遷移學習函數(shù)得到兩種特征,再加上古代詩歌本身的特征,建立3個分類器并投票得出最后的結果。在中國唐詩上的實驗表明了方法的有效性,并詳細分析了唐代各個時期以及重要流派的情感,結合社會歷史的分析,取得了良好的效果。
[Abstract]:With the development of Web 2.0, people have been used to express their views and opinions on the Internet, and to obtain the information they need from the information published by others, thus forming an Internet model dominated by the vast number of users. Under this kind of Internet model, people rely more and more on the Internet. From the initial search for information to the various chat communities later, to the present, they have to go to the Internet to see other people's comments before they can make a decision. The Internet is changing every aspect of people's lives. Social media is the medium of these behaviors, including virtual communities and web platforms, where people can create, share, exchange opinions, ideas and experiences, including Weibo, blogs, forums, online communities, comment sites and so on. People express their views on social media, and personal views tend to have emotional preferences, which fall into two categories: text information, such as the content of Weibo, and scoring information. Such as the film score, and so on. User preference refers to the user's affection for an event, object, disgust and so on. The research of user preference is to understand the emotional preference that users want to express by studying the rich emotional information. This paper will study the problem of user preference in social media from two aspects: score prediction of aspect score and emotional analysis of Tang poetry. The aspect score is the score of all aspects of the product, while the total score is the comprehensive score of all aspects of the product. Nowadays, most of the work using the total score is based on the assumption that the total score is the average score of the aspect score or the total score is very close to the aspect score. However, after analyzing the real data set, it is found that there is a score deviation between the total score and the aspect score, but the existing work does not consider the score deviation. In this paper, the problem of aspect score prediction with score deviation is studied for the first time, and a new affective subject hybrid model, RCMB.RCMB, is proposed, which considers that the total score is the center of the probability graph and integrates the prior information of the score deviation through a hidden aspect scoring variable. Experiments on real data sets (Dianping and TripAdvisor) show that RCMB has higher prediction accuracy than other existing methods and is more able to maintain the relative order of scores. Current affective analysis works generally focus on modern texts, such as product reviews and Weibo, with little reference to the analysis of ancient literature. Poetry is the same as Weibo used by ancient people, and it is also an important medium to express their feelings. This paper presents an emotional classification model of Chinese Tang poetry based on transfer learning, TL-PCO. by analyzing the emotion of poetry, we can understand the social and cultural progress at that time. TL-PCO obtains two characteristics through two transfer learning functions. Combined with the characteristics of ancient poetry, three classifiers were established and voted for the final result. The experiment in Chinese Tang poetry shows the effectiveness of the method, and analyzes the emotion of every period and important schools in the Tang Dynasty in detail, combining with the analysis of social history, it has achieved good results.
【學位授予單位】：北京郵電大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TP391.3
本文編號：2157911
[Abstract]:With the development of Web 2.0, people have been used to express their views and opinions on the Internet, and to obtain the information they need from the information published by others, thus forming an Internet model dominated by the vast number of users. Under this kind of Internet model, people rely more and more on the Internet. From the initial search for information to the various chat communities later, to the present, they have to go to the Internet to see other people's comments before they can make a decision. The Internet is changing every aspect of people's lives. Social media is the medium of these behaviors, including virtual communities and web platforms, where people can create, share, exchange opinions, ideas and experiences, including Weibo, blogs, forums, online communities, comment sites and so on. People express their views on social media, and personal views tend to have emotional preferences, which fall into two categories: text information, such as the content of Weibo, and scoring information. Such as the film score, and so on. User preference refers to the user's affection for an event, object, disgust and so on. The research of user preference is to understand the emotional preference that users want to express by studying the rich emotional information. This paper will study the problem of user preference in social media from two aspects: score prediction of aspect score and emotional analysis of Tang poetry. The aspect score is the score of all aspects of the product, while the total score is the comprehensive score of all aspects of the product. Nowadays, most of the work using the total score is based on the assumption that the total score is the average score of the aspect score or the total score is very close to the aspect score. However, after analyzing the real data set, it is found that there is a score deviation between the total score and the aspect score, but the existing work does not consider the score deviation. In this paper, the problem of aspect score prediction with score deviation is studied for the first time, and a new affective subject hybrid model, RCMB.RCMB, is proposed, which considers that the total score is the center of the probability graph and integrates the prior information of the score deviation through a hidden aspect scoring variable. Experiments on real data sets (Dianping and TripAdvisor) show that RCMB has higher prediction accuracy than other existing methods and is more able to maintain the relative order of scores. Current affective analysis works generally focus on modern texts, such as product reviews and Weibo, with little reference to the analysis of ancient literature. Poetry is the same as Weibo used by ancient people, and it is also an important medium to express their feelings. This paper presents an emotional classification model of Chinese Tang poetry based on transfer learning, TL-PCO. by analyzing the emotion of poetry, we can understand the social and cultural progress at that time. TL-PCO obtains two characteristics through two transfer learning functions. Combined with the characteristics of ancient poetry, three classifiers were established and voted for the final result. The experiment in Chinese Tang poetry shows the effectiveness of the method, and analyzes the emotion of every period and important schools in the Tang Dynasty in detail, combining with the analysis of social history, it has achieved good results.
【學位授予單位】：北京郵電大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TP391.3
【參考文獻】
相關期刊論文 前1條
1 劉巖斌，俞士汶，，孫欽善;古詩研究的計算機支持環(huán)境的實現(xiàn)[J];中文信息學報;1997年01期
相關博士學位論文 前1條
1 常娥;古籍智能處理技術研究[D];南京農業(yè)大學;2007年
相關碩士學位論文 前1條
1 葉振超;CADAL中國文學編年史系統(tǒng)的設計與實現(xiàn)[D];浙江大學;2011年
本文編號：2157911
本文鏈接：http://sikaile.net/kejilunwen/ruanjiangongchenglunwen/2157911.html
最近更新
教材專著
