本文選題：用戶瀏覽行為 + 用戶興趣��； 參考：《東北大學(xué)》2013年碩士論文

【摘要】：隨著Web上資源的急劇膨脹,面對(duì)用戶提供的有限查詢?cè)~,當(dāng)前的搜索引擎提供的千人一面的搜索已經(jīng)難以滿足用戶對(duì)搜索結(jié)果的需求。在用戶使用搜索引擎進(jìn)行信息檢索的過(guò)程中,依據(jù)用戶的實(shí)際興趣為用戶返回個(gè)性化的搜索結(jié)果可以提高用戶對(duì)搜索結(jié)果的滿意度。大量研究表明用戶的實(shí)際興趣與其在網(wǎng)頁(yè)上的瀏覽行為是密切相關(guān)的,通過(guò)用戶瀏覽行為分析可以獲取用戶興趣信息,進(jìn)而構(gòu)建用戶興趣模型,使搜索結(jié)果更加貼近用戶的期望。然而,目前的隱式用戶興趣獲取方法無(wú)法很好的預(yù)測(cè)出用戶對(duì)頁(yè)面的實(shí)際興趣度。究其原因,一方面是由于當(dāng)前研究尚未考慮到用戶的瀏覽行為可能隨搜索任務(wù)類型的不同而變化。另一方面,當(dāng)前的用戶興趣獲取方法多使用某種特定用戶行為預(yù)測(cè)用戶興趣度。 針對(duì)上述問(wèn)題,本文探究用戶瀏覽行為在不同類型的搜索任務(wù)中所表現(xiàn)出的差異,并研究聯(lián)合分析多種用戶瀏覽行為的隱式用戶興趣獲取方法。在此基礎(chǔ)上構(gòu)建適當(dāng)用戶興趣模型,最終得出用戶的實(shí)際興趣,從而實(shí)現(xiàn)個(gè)性化服務(wù),使搜索結(jié)果更加貼近用戶的期望。 具體的,本文將任務(wù)類型分為導(dǎo)航型、信息型、事務(wù)型三種不同類型,將用戶的基本瀏覽行為轉(zhuǎn)換為頁(yè)面停留時(shí)間時(shí)間、鼠標(biāo)點(diǎn)擊次數(shù)、頁(yè)面重訪問(wèn)次數(shù)以及滑塊移動(dòng)次數(shù)四種可分析行為事件。通過(guò)Bernard提出的算法完成了任務(wù)類型的自動(dòng)識(shí)別,分析了四種可分析行為事件在不同搜索任務(wù)類型中表現(xiàn)出的差異。在用戶行為分析階段,本文基于M5模型樹對(duì)可分析事件建模完成對(duì)用戶興趣度的計(jì)算,在計(jì)算過(guò)程中樹的剪枝和相關(guān)系數(shù)平滑是建模過(guò)程中必須考慮的問(wèn)題。模型評(píng)價(jià)階段,本文使用模型準(zhǔn)確率評(píng)價(jià)指標(biāo)將不區(qū)分任務(wù)類型和區(qū)分任務(wù)類型的模型與Nicholas Belkin的模型進(jìn)行了對(duì)比。為了清晰有效的表達(dá)用戶興趣信息,本文提出了基于分類的用戶興趣模型,該模型涉及對(duì)文檔的特征值提取,基于搜狗語(yǔ)料的SVM分類器對(duì)相關(guān)文檔進(jìn)行分類等技術(shù)。使用準(zhǔn)確率和排序準(zhǔn)確率兩個(gè)指標(biāo)將baidu搜索引擎和基于VSM的模型及基于分類的模型進(jìn)行了對(duì)比。實(shí)驗(yàn)結(jié)果表明,本文提出的面向用戶興趣的用戶行為分析模型可有效提高用戶對(duì)搜索結(jié)果的滿意度。
[Abstract]:With the rapid expansion of resources on Web, facing the limited query words provided by users, the current search engine can not meet the demand of search results. In the process of information retrieval by using search engine, the user's satisfaction with search results can be improved by returning personalized search results to users according to their actual interests. A large number of studies show that the actual interest of users is closely related to their browsing behavior on the web. Through the analysis of user browsing behavior, user interest information can be obtained and then user interest models can be constructed. Make search results closer to user expectations. However, the current implicit user interest acquisition method can not predict the actual user interest in the page. On the one hand, the current research has not considered that the browsing behavior of users may vary with the type of search task. On the other hand, the current user interest acquisition methods often use a specific user behavior to predict user interest. To solve the above problems, this paper explores the differences of user browsing behavior in different types of search tasks, and studies an implicit user interest acquisition method which can jointly analyze multiple user browsing behaviors. On this basis, the appropriate user interest model is constructed, and finally the actual interest of the user is obtained, thus the personalized service is realized, and the search results are closer to the user's expectation. Specifically, the task type is divided into three types: navigation type, information type and transaction type. The basic browsing behavior of the user is converted into page stay time, mouse click times, etc. Page revisits and slider moves four analyzable behavior events. The automatic recognition of task types is accomplished by the algorithm proposed by Bernard, and the differences of four analyzable behavior events in different search task types are analyzed. In the phase of user behavior analysis, the user interest is calculated based on M5 model tree. In the process of computing, the tree pruning and correlation coefficient smoothing are the problems that must be considered in the modeling process. In the stage of model evaluation, the model that does not distinguish between task type and task type is compared with that of Nicholas Belkin. In order to express user interest information clearly and effectively, this paper proposes a classification-based user interest model, which involves the extraction of feature values of documents and the classification of relevant documents by SVM classifier based on Sogou corpus. The baidu search engine is compared with the model based on VSM and the model based on classification by using accuracy and sorting accuracy. The experimental results show that the user behavior analysis model proposed in this paper can effectively improve the users' satisfaction with the search results.
【學(xué)位授予單位】：東北大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：TP391.3

【參考文獻(xiàn)】

中國(guó)期刊全文數(shù)據(jù)庫(kù) 前1條

1 王川;王大玲;于戈;馬海濤;劉鑫鋼;;基于用戶行為模型的搜索引擎[J];計(jì)算機(jī)工程;2008年04期

，

本文編號(hào)：1814889