發(fā)布時間：2018-03-18 20:17

  本文選題：用戶興趣　切入點：LDA　出處：《北京郵電大學》2017年碩士論文　論文類型：學位論文

【摘要】：以微博為代表的社交網(wǎng)絡平臺在日常生活中越來越活躍。微博平臺上眾多微博用戶實時產(chǎn)生的海量微博內(nèi)容,帶來了信息冗余的問題,給用戶的使用體驗帶來挑戰(zhàn)。發(fā)現(xiàn)微博用戶個人喜好和興趣,為用戶帶來效率更高、更精確的使用體驗;識別微博用戶的行為意圖,為微博營銷平臺提供更加精準的指導,是微博研究中兩個重大問題。本文較為深入地研究這兩個問題,提出了解決方案。發(fā)現(xiàn)微博用戶的興趣,面對的主要問題是微博文本的短文本特性。傳統(tǒng)文本處理方法在特征稀疏、微博文本上下文依賴性問題上有著局限性。本文基于傳統(tǒng)LDA特征拓展方法進行改進,將文本-主題分布中的文本主題特征引入文本特征空間,進一步拓展特征。利用主題模型識別微博中的多義詞,消除多義詞影響,進一步提升分類算法的性能。通過實驗來驗證方法的有效性。識別微博用戶的意圖,是微博研究中較為新穎的領(lǐng)域。在本研究中,先提取潛在的意圖微博,根據(jù)意圖微博特征,將包含意圖指示詞的微博提取出來。然后根據(jù)圖傳播模型對微博進行意圖分類。并通過實驗來驗證該方法的有效性。
[Abstract]:The social network platform, represented by Weibo, is becoming more and more active in daily life. The huge amount of Weibo content generated in real time by a large number of Weibo users on the Weibo platform has brought about the problem of information redundancy. It brings challenges to the user's use experience. It is found that Weibo's personal preferences and interests bring users a more efficient and accurate use experience; identify the user's behavior intention; and provide more precise guidance for Weibo's marketing platform. It is two major problems in Weibo's research. This paper studies these two problems in depth and puts forward a solution. We find the interest of Weibo users. The main problem is the short text feature of Weibo text. The traditional text processing method has some limitations in feature sparsity, Weibo text context-dependent problem. This paper improves on the traditional LDA feature extension method. The text theme features in the text-theme distribution are introduced into the text feature space to further expand the features. The polysemous words in Weibo are identified by the thematic model to eliminate the influence of polysemous words. To further improve the performance of the classification algorithm. Through experiments to verify the effectiveness of the method. Identification of Weibo user intention, is a relatively new area of research. In this study, the potential intention of Weibo is extracted, according to the characteristics of the intention Weibo, Weibo was extracted from the deixis of intention. Then the intention was classified according to the graph propagation model, and the validity of the method was verified by experiments.
【學位授予單位】：北京郵電大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP393.092;TP391.1

【參考文獻】

相關(guān)期刊論文 前2條

1 彭敏;黃佳佳;朱佳暉;黃濟民;劉紀平;;基于頻繁項集的海量短文本聚類與主題抽取[J];計算機研究與發(fā)展;2015年09期

2 翟延冬;王康平;張東娜;黃嵐;周春光;;一種基于WordNet的短文本語義相似性算法[J];電子學報;2012年03期

，

本文編號：1631140