面向IPTV的混合式自適應(yīng)推薦系統(tǒng)關(guān)鍵技術(shù)研究與實現(xiàn)
發(fā)布時間：2018-10-23 10:41
【摘要】： IPTV作為新一代有線數(shù)字電視產(chǎn)品,自從進(jìn)入中國以來,用戶量增長迅速。根據(jù)權(quán)威機構(gòu)IDC的預(yù)測,到2009年底,中國IPTV用戶量將達(dá)到460萬,而到2013年,這數(shù)字將增長到1310萬,并將進(jìn)入一個井噴式的發(fā)展。 IPTV的主要優(yōu)勢在于其良好的互動性。通過IPTV,用戶將在“IP機頂盒+電視機”上告別單一被動的節(jié)目接收,走向更為豐富多彩的互動數(shù)字娛樂生活。內(nèi)容服務(wù)提供商可以在IPTV上提供大量高質(zhì)量的數(shù)字圖像、視頻、音頻、游戲、遠(yuǎn)程教育、廣告等內(nèi)容。在這種環(huán)境下,大量的信息容易讓用戶產(chǎn)生信息迷失。因此為用戶提供精準(zhǔn)高質(zhì)的個性化服務(wù)成為一種迫切的需求。目前世界范圍內(nèi)對個性化服務(wù)的研究主要歸為對推薦系統(tǒng)的研究范疇。 文章首先深入分析現(xiàn)有推薦系統(tǒng)算法所存在的不足,其中包括新用戶問題以及混合式過濾算法所采用的固定混合比造成的推薦質(zhì)量下降等問題。作者針對這些問題展開研究討論。 首先針對新用戶問題,文章提出了基于人口屬性的協(xié)作過濾算法,這個算法將人口屬性信息相似度引入?yún)f(xié)作過濾算法,并和PCC計算所得相似度進(jìn)行混合得到新的相似度。采用這個相似度計算最近鄰并產(chǎn)生推薦。實驗分析表明,文章提出的基于人口屬性的協(xié)作過濾在用戶評分稀少,用戶profile稀疏的時候能夠有效提高推薦質(zhì)量。 之后針對傳統(tǒng)混合式推薦系統(tǒng)造成推薦質(zhì)量下降問題,提出了基于遞度下降的混合式自適應(yīng)推薦算法。本算法引入自學(xué)習(xí)機制,讓系統(tǒng)自動調(diào)整混合式推薦系統(tǒng)的混合比。實驗表明,這個算法在一定程度上提高了推薦精度,并且不增加過多的額外計算時間。 文章的第三個成果是通過對IPTV平臺特性的分析,以及將它同現(xiàn)有以個人電腦為終端的推薦系統(tǒng)的比較,總結(jié)出面向IPTV的推薦系統(tǒng)所應(yīng)該具有的特性:用戶零學(xué)習(xí)成本、用戶零額外操作。針對這個特性,文章設(shè)計了一個用戶喜好挖掘算法,通過分析用戶的訪問日志,自動獲取用戶的喜好。經(jīng)過系統(tǒng)一年的線上運行,證明此算法運行效果良好。
[Abstract]:IPTV as a new generation of cable digital television products, since entering China, the number of users growing rapidly. According to IDC, an authoritative organization, the number of IPTV users in China will reach 4.6 million by the end of 2009, and will increase to 13.1 million by 2013. And will enter a blowout development. The main advantage of IPTV lies in its good interactivity. IPTV, users will bid farewell to the single passive program reception on "IP set-top box TV" and move towards a more colorful interactive digital entertainment life. Content service providers can provide high-quality digital images, video, audio, games, distance education, advertising and other content on IPTV. In this environment, a large number of information is easy to make users lose information. Therefore, it is an urgent need to provide users with accurate and high quality personalized services. At present, the research on personalized service is classified as recommendation system. Firstly, this paper deeply analyzes the shortcomings of the existing recommendation system algorithms, including the problem of new users and the degradation of recommendation quality caused by the fixed mixing ratio of hybrid filtering algorithm. The author studies and discusses these problems. To solve the problem of new users, a collaborative filtering algorithm based on population attributes is proposed in this paper. This algorithm introduces the similarity of population attributes into the collaborative filtering algorithm, and mixes with the similarity calculated by PCC to obtain a new similarity. This similarity is used to calculate the nearest neighbor and produce recommendations. The experimental results show that the proposed collaborative filtering based on population attributes can effectively improve the recommendation quality when the user score is scarce and the user profile is sparse. After that, a hybrid adaptive recommendation algorithm based on transitivity reduction is proposed to solve the problem of the deterioration of recommendation quality caused by the traditional hybrid recommendation system. This algorithm introduces self-learning mechanism and allows the system to automatically adjust the hybrid ratio of hybrid recommendation systems. Experiments show that the proposed algorithm improves the recommendation accuracy to some extent and does not increase the extra computation time. The third result of this paper is to analyze the characteristics of IPTV platform and compare it with the existing recommendation system with personal computer as the terminal, and summarize the characteristics of the recommendation system to IPTV: user zero learning cost. User zero extra operation. Aiming at this feature, this paper designs a user preference mining algorithm, which automatically acquires user preferences by analyzing the user's access log. It is proved that the algorithm works well after one year's running.
【學(xué)位授予單位】：華東師范大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2010
【分類號】：TN949.2
本文編號：2288994
[Abstract]:IPTV as a new generation of cable digital television products, since entering China, the number of users growing rapidly. According to IDC, an authoritative organization, the number of IPTV users in China will reach 4.6 million by the end of 2009, and will increase to 13.1 million by 2013. And will enter a blowout development. The main advantage of IPTV lies in its good interactivity. IPTV, users will bid farewell to the single passive program reception on "IP set-top box TV" and move towards a more colorful interactive digital entertainment life. Content service providers can provide high-quality digital images, video, audio, games, distance education, advertising and other content on IPTV. In this environment, a large number of information is easy to make users lose information. Therefore, it is an urgent need to provide users with accurate and high quality personalized services. At present, the research on personalized service is classified as recommendation system. Firstly, this paper deeply analyzes the shortcomings of the existing recommendation system algorithms, including the problem of new users and the degradation of recommendation quality caused by the fixed mixing ratio of hybrid filtering algorithm. The author studies and discusses these problems. To solve the problem of new users, a collaborative filtering algorithm based on population attributes is proposed in this paper. This algorithm introduces the similarity of population attributes into the collaborative filtering algorithm, and mixes with the similarity calculated by PCC to obtain a new similarity. This similarity is used to calculate the nearest neighbor and produce recommendations. The experimental results show that the proposed collaborative filtering based on population attributes can effectively improve the recommendation quality when the user score is scarce and the user profile is sparse. After that, a hybrid adaptive recommendation algorithm based on transitivity reduction is proposed to solve the problem of the deterioration of recommendation quality caused by the traditional hybrid recommendation system. This algorithm introduces self-learning mechanism and allows the system to automatically adjust the hybrid ratio of hybrid recommendation systems. Experiments show that the proposed algorithm improves the recommendation accuracy to some extent and does not increase the extra computation time. The third result of this paper is to analyze the characteristics of IPTV platform and compare it with the existing recommendation system with personal computer as the terminal, and summarize the characteristics of the recommendation system to IPTV: user zero learning cost. User zero extra operation. Aiming at this feature, this paper designs a user preference mining algorithm, which automatically acquires user preferences by analyzing the user's access log. It is proved that the algorithm works well after one year's running.
【學(xué)位授予單位】：華東師范大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2010
【分類號】：TN949.2
【參考文獻(xiàn)】
相關(guān)期刊論文 前2條
1 曾春,邢春曉,周立柱;個性化服務(wù)技術(shù)綜述[J];軟件學(xué)報;2002年10期
2 曾春,邢春曉,周立柱;基于內(nèi)容過濾的個性化搜索算法[J];軟件學(xué)報;2003年05期
，本文編號：2288994
本文鏈接：http://www.lk138.cn/wenyilunwen/guanggaoshejilunwen/2288994.html
最近更新
教材專著
