【摘要】： 文本自動(dòng)分類是指在給定的分類體系下,根據(jù)文本內(nèi)容自動(dòng)確定文本所屬類別。文本分類技術(shù)的出現(xiàn),使文檔可以自動(dòng)地按照類別組織和處理,符合人類組織和處理信息的方式。同時(shí),作為信息過濾、信息檢索、搜索引擎等領(lǐng)域的技術(shù)基礎(chǔ),文本分類技術(shù)有著廣泛的應(yīng)用前景。 學(xué)報(bào)和學(xué)術(shù)會議所使用的論文投稿系統(tǒng),涉及上千篇投稿論文要分配給上百位評審專家去審閱,在很短的時(shí)間內(nèi)人工分配這些投稿論文給相關(guān)學(xué)科領(lǐng)域的專家們?nèi)ピu審?fù)ヅ涞牟缓�。特別是評審專家的研究領(lǐng)域不清楚,人工無法及時(shí)、準(zhǔn)確的收集到評審專家所屬的學(xué)科領(lǐng)域信息,影響到論文分配任務(wù)的正常進(jìn)行。選擇合適的評審專家是正確評價(jià)投稿論文質(zhì)量和提升學(xué)報(bào)、期刊學(xué)術(shù)層次的關(guān)鍵,如何用計(jì)算機(jī)來實(shí)現(xiàn)自動(dòng)分配投稿論文給匹配領(lǐng)域的評審專家去審閱?文本自動(dòng)分類可以很好的解決這個(gè)問題。 論文針對上述問題,提出一種基于文本分類技術(shù)的評審專家自動(dòng)推薦模型,通過文本分類技術(shù)對投稿論文和對評審專家所發(fā)表的論文進(jìn)行所屬學(xué)科領(lǐng)域的分類,進(jìn)而判斷出評審專家的主要研究領(lǐng)域和投稿論文的學(xué)科領(lǐng)域。然后將投稿論文的學(xué)科領(lǐng)域與評審專家的研究領(lǐng)域自動(dòng)匹配,建立自動(dòng)推薦評審專家模型。論文的主要研究內(nèi)容如下: ①在特征篩選中,引入最大頻率的概念和特征項(xiàng)與類別的相關(guān)系數(shù)D ( m_(ik)),提出了改進(jìn)的χ~2算法,實(shí)驗(yàn)結(jié)果表明,在特征項(xiàng)篩選中表現(xiàn)出了良好的篩選效果。 ②針對評審專家自動(dòng)推薦模型選取的特征項(xiàng)為論文的關(guān)鍵詞,在文本向量表示方法的基礎(chǔ)上作了簡化,提出了基于TF/IDF特征權(quán)重閾值的向量空間模型算法,并選用SVM分類方法對特征矩陣分類。實(shí)驗(yàn)結(jié)果表明,該算法可以有效的濾除不相關(guān)的噪聲特征,產(chǎn)生更為準(zhǔn)確的分類模型。 ③針對主動(dòng)學(xué)習(xí)SVM分類算法在多類別的分類問題上存在分類器的速度隨數(shù)目增加而變慢的問題,引入有向無環(huán)圖SVM,改進(jìn)了主動(dòng)學(xué)習(xí)SVM分類算法,實(shí)驗(yàn)結(jié)果表明,改進(jìn)后主動(dòng)學(xué)習(xí)SVM分類算法可以增加交互的過程使訓(xùn)練得到的分類器具備自學(xué)習(xí)的能力,改進(jìn)后主動(dòng)學(xué)習(xí)SVM分類器在多類別的分類上能夠精確分類并且提高分類速度。
[Abstract]:Automatic text categorization refers to the automatic classification of text according to the text content under a given classification system. The emergence of text categorization technology enables documents to be organized and processed automatically according to categories, in line with the way human beings organize and process information. At the same time, as the technical foundation of information filtering, information retrieval, search engine and other fields, text classification technology has a wide application prospect. The paper submission system used in journals and academic conferences involves the assignment of thousands of papers to hundreds of experts to review and review them. The manual assignment of these contributions to experts in related disciplines in a short period of time is often poorly matched. Especially, the research field of the review experts is not clear, and people can not collect the information of the subject fields of the review experts in time and accurately, which affects the normal work of the assignment of papers. It is the key to correctly evaluate the quality of contribution papers and improve the academic level of journals to select suitable review experts. How to use computer to automatically assign contribution papers to the evaluation experts in the matching field to review? Automatic text categorization can solve this problem very well. In order to solve the above problems, an automatic recommendation model of review experts based on text classification technology is put forward in this paper, which classifies the contribution papers and the papers published by the review experts through the text classification technology. The main research field and the subject field of the contribution paper are judged. Then the subject field of the contribution paper is automatically matched with the research field of the evaluation expert, and the expert model of automatic recommendation and evaluation is established. The main contents of this paper are as follows: 1 in feature selection, the concept of maximum frequency and the correlation coefficient D (m _ (ik),) are introduced. The experimental results show that, It shows good screening effect in feature selection. (2) aiming at the feature items selected from the automatic recommendation model of evaluation experts as the key words of the paper, the text vector representation method is simplified, and a vector space model algorithm based on TF/IDF feature weight threshold is proposed. SVM classification method is used to classify the feature matrix. Experimental results show that the algorithm can effectively filter irrelevant noise features and produce more accurate classification models. 3 in order to solve the problem that the speed of classifier becomes slower with the increase of the number of classifiers in the multi-class SVM classification algorithm, the active learning SVM classification algorithm is improved by introducing the directed acyclic graph SVM,. The improved active learning SVM classification algorithm can increase the interactive process so that the trained classifier has the ability of self-learning. The improved active learning SVM classifier can accurately classify and improve the classification speed in multi-category classification.
【學(xué)位授予單位】：重慶大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2009
【分類號】：TP182

【引證文獻(xiàn)】

相關(guān)博士學(xué)位論文 前1條

1 向東;產(chǎn)品設(shè)計(jì)中多領(lǐng)域知識表達(dá)、獲取及應(yīng)用研究[D];華中科技大學(xué);2012年

，

本文編號：2339205