本文選題：用戶偏好 + 時(shí)間動(dòng)態(tài)　； 參考：《清華大學(xué)》2015年碩士論文

【摘要】：在推薦系統(tǒng)中,用戶對(duì)產(chǎn)品的偏好往往會(huì)隨著時(shí)間而發(fā)生動(dòng)態(tài)的變化。從用戶個(gè)人的角度出發(fā),這意味著用戶在不斷與產(chǎn)品相接觸的過程中積累了越來越多的知識(shí)與經(jīng)驗(yàn),從“新手”不斷地向“專家”發(fā)展,從而使自己具有了不同的品位。例如,一位電影領(lǐng)域的“新手”用戶可能會(huì)認(rèn)為《黑客帝國》甚至有些無聊,而當(dāng)他看過許多電影之后重新審視這部電影時(shí),才能更好地理解并欣賞這部電影,從而給出更高的評(píng)價(jià)。同時(shí),用戶在推薦平臺(tái)上給出評(píng)分時(shí),可能會(huì)受到兩種因素的即時(shí)性影響,分別為產(chǎn)品的平均得分與之前其他若干用戶對(duì)產(chǎn)品的評(píng)分。例如,若某一用戶試圖為某一產(chǎn)品給出5分,但當(dāng)他發(fā)現(xiàn)該產(chǎn)品的平均得分不到3分,或者在他之前的若干用戶給出了1~2分的低分時(shí),他可能會(huì)降低自己所給出的分?jǐn)?shù)。為了掌握用戶經(jīng)驗(yàn)水平的發(fā)展模式,同時(shí)發(fā)現(xiàn)用戶在評(píng)分時(shí)受到其他用戶影響的機(jī)制,本文主要研究如何將隱馬爾可夫模型與協(xié)同過濾算法相結(jié)合,以提升推薦系統(tǒng)的準(zhǔn)確度,發(fā)現(xiàn)用戶的行為模式,為電子商務(wù)中的精準(zhǔn)營銷提供理論支持。首先,在總結(jié)前人工作的基礎(chǔ)上,基于所要解決問題的特征,本文提出了將隱馬爾可夫模型與協(xié)同過濾算法相結(jié)合,來描述用戶經(jīng)驗(yàn)等級(jí)發(fā)展的模型。此后,在此模型的基礎(chǔ)上,又添加了描述其他用戶影響的參數(shù),旨在研究用戶在評(píng)分時(shí)受到其他用戶影響的機(jī)制。同時(shí),本文在四個(gè)推薦系統(tǒng)相關(guān)的數(shù)據(jù)集上進(jìn)行了實(shí)驗(yàn)。實(shí)驗(yàn)結(jié)果表明,本文所提出的模型在預(yù)測準(zhǔn)確率上優(yōu)于前人所提出的模型。同時(shí),通過對(duì)模型優(yōu)化與參數(shù)學(xué)習(xí)結(jié)果的分析,也發(fā)現(xiàn)了一些有意義的結(jié)論。本研究在提升預(yù)測準(zhǔn)確率的同時(shí),在一定程度上優(yōu)化了前人工作中所存在的一些不足,同時(shí)將隱馬爾可夫模型應(yīng)用于推薦系統(tǒng)領(lǐng)域,在理論上豐富了推薦系統(tǒng)的算法。同時(shí),本文所進(jìn)行的對(duì)于用戶偏好動(dòng)態(tài)的研究以及對(duì)于用戶經(jīng)驗(yàn)等級(jí)發(fā)展的建模機(jī)制,對(duì)于多指標(biāo)推薦問題、專家發(fā)現(xiàn)與專家推薦問題等也具有一定的啟示意義。
[Abstract]:In recommendation systems, users' preferences for products often change dynamically over time. From the individual point of view of the user, this means that the user has accumulated more and more knowledge and experience in the process of continuous contact with the product, and has developed from "novice" to "expert", so that he has different taste. For example, a "novice" user of the film industry may think the Matrix is even boring, and when he has seen many movies and revisited the movie, he can better understand and appreciate the film. Thus giving a higher evaluation. At the same time, the users may be affected by two kinds of factors, which are the average score of the product and the score of the other users when they give the rating on the recommendation platform. For example, if a user tries to give a product a score of 5, but finds that the average score of the product is less than 3, or several users before him give a low score of 1 to 2, he may reduce his score. In order to grasp the development mode of user experience level and find out the mechanism that users are influenced by other users in scoring, this paper mainly studies how to combine hidden Markov model with collaborative filtering algorithm to improve the accuracy of recommendation system. Discover the user's behavior pattern and provide theoretical support for accurate marketing in e-commerce. First of all, based on the previous work and the characteristics of the problem to be solved, this paper proposes a model that combines hidden Markov model with collaborative filtering algorithm to describe the development of user experience level. Then, on the basis of this model, parameters describing the influence of other users are added to study the mechanism by which users are influenced by other users. At the same time, the experiment is carried out on the data sets of four recommendation systems. The experimental results show that the prediction accuracy of the proposed model is better than that of the previous model. At the same time, by analyzing the results of model optimization and parameter learning, some meaningful conclusions are also found. This study not only improves the accuracy of prediction, but also optimizes some shortcomings of previous work to some extent. At the same time, the hidden Markov model is applied to the field of recommendation system, which enriches the algorithm of recommendation system in theory. At the same time, the research on user preference dynamics, the modeling mechanism for the development of user experience level, the multi-index recommendation problem, the expert discovery and the expert recommendation problem also have certain enlightenment significance.
【學(xué)位授予單位】：清華大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：F724.6

【引證文獻(xiàn)】

相關(guān)碩士學(xué)位論文 前1條

1 成偉丹;基于遺忘函數(shù)和用戶的協(xié)同過濾推薦算法研究[D];浙江工業(yè)大學(xué);2016年

，

本文編號(hào)：1962665