臺灣博碩士論文加值系統

在現今網路行銷的時代，網路廣告是一個很重要的行銷手法。本論文提出一個運用關鍵字與使用者輪廓開發一個以本體論為基礎的網路廣告推薦系統之流程與架構，其主要目的是推薦使用者其有興趣之廣告，首先我們參考一些專家的分類方式(如Yahoo關鍵字廣告)，來將廣告分類，並經由斷詞系統(CKIP)取得關鍵字，接著利用詞彙頻率與反向文件頻率法(TF-IDF, Term Frequency–Inverse Document Frequency)，找出數個能代表每一類別的代表關鍵字，以此分析並建立廣告的關鍵字與廣告類別之關係，建構一本體論(Ontology)來描述之。再建立一廣告清單供使用者選擇其感興趣的廣告，透過收集分析使用者的回應，計算所選廣告的關鍵字與本體論(Ontology)中每個類別的代表關鍵字間的NGD (Normalized Google Distance)值，來得出相似度並構成每個廣告的特徵值(Characteristic Vector)，以此視為使用者的長期興趣(Long-term Interests)；而當使用者進行搜尋時，記錄使用者搜尋當下輸入的文字，以斷詞系統找出關鍵字，再用NGD找出特徵值，以此視為使用者的短期興趣(Short-term Interests)。將以上長期與短期興趣得出的兩組特徵值，運用邏輯式迴歸分析(Logistic Regression)來建立使用者特性(User Profile)，如此一來就可以根據使用者特性與當下的喜好與廣告的特徵值來計算得出一個推薦參考值，利用這個參考值即可來判斷推薦使用者有興趣之廣告。因此我們開發一個推薦系統，來推薦使用者有興趣而會去點選瀏覽的網路廣告。實驗證明，相較於隨機推薦，所提出的方法可以顯著地提升推薦成功的比率，因此將可提升網路行銷的效益。

This thesis focused on developing a personalized internet advertisement recommendation system. First, all the advertisements are categorized by referring to the commercial categories provided by Yahoo. Then, the representative keywords of each category are extracted from each advertisement and are identified through Term Frequency and Inverted Domain Frequency (TF-IDF) method. Thus, the ontology of the advertisements is built. Second, Normalized Google Distance (NGD) values between keywords are computed to derive the characteristic vector of each advertisement. Hence, the so-called characteristic vector of each advertisement is established. Also, the user profile, which describes a user’s preferences, is established based on the user’s long-term and short-term interests through Logistic Regression. Finally, for a new advertisement, by using the characteristic vector of this advertisement and the user profile, a recommendation value can be computed. This value is used to determine whether this advertisement should be recommended to the user or not. A prototype website was developed for verifying the proposed schemes. Our experiment results showed that the proposed personalized advertisement recommendation system has better performance than the random selection methods.

Table of Contents
摘要---------------------------------------------------- i
Abstract----------------------------------------------- ii
Chapter 1. Introduction-------------------------------- 1
1.1 Background------------------------------------- 1
1.2 Motivation and Challenges---------------------- 1
1.3 Solutions and Contributions-------------------- 3
1.4 Thesis Organization---------------------------- 5
Chapter 2. Related Work-------------------------------- 6
2.1 Personalized Advertisement Recommendation---------- 6
2.2. TF-IFD (Term Frequency–Inverse Document Frequency) 6
2.3. Ontology------------------------------------------ 8
2.4. NGD (Normalized Google Distance)------------------ 9
2.5. Logistic Regression------------------------------- 10
Chapter 3. The Proposed Approach----------------------- 13
3.1. Architecture of the Proposed System--------------- 13
3.2. Implementation Details---------------------------- 15
3.2.1 Building Ontology-------------------------------- 15
3.2.2 Advertisement Characteristic Vector-------------- 18
3.2.3 Collecting Users’ Interests---------------------- 19
3.2.4 User Profile Establishment----------------------- 21
3.2.5 Advertisement Recommendation--------------------- 23
Chapter 4. Experiments and Evaluation------------------ 24
4.1. Experiments--------------------------------------- 24
4.2. Evaluation---------------------------------------- 29
Chapter 5. Conclusions--------------------------------- 32
References--------------------------------------------- 33


List of Tables
Table 1. Outcome of the recommendation.---------------- 30

 
List of Figures
Figure 1. The standard logistic regression function---- 11
Figure 2. System Architecture (a) Advertisements Ontology (b) User Profile (c) Advertisement Recommendation----------------------- 15
Figure 3. Advertisement categorization provided by Yahoo------- 16
Figure 4. Example of an Advertisement Ontology--------- 17
Figure 5. Word segmentation using CKIP----------------- 25
Figure 6. The advertisement ontology.------------------ 26
Figure 7. 50 advertisements are randomly chosen to collect users’ preferences-------------------------------------------- 28
Figure 8. Recommend advertisements--------------------- 29
Figure 9. The results of SPSS T-test------------------- 31

References
[1]Van Meteren R., and Van Someren M., “Using Content-Based Filtering for Recommendation,” Proceedings of the Machine Learning in the New Information Age: MLnet/ECML2000 Workshop,pp. 47-56, 2000.
[2]Linden Greg, Brent Smith, and Jeremy York, “Amazon. com recommendations: Item-to-item collaborative filtering,” Internet Computing, IEEE 7(1), pp. 76-80, 2003.
[3]Schafer J. Ben, Joseph A. Konstan, and John Riedl, “E-commerce recommendation applications,” Applications of Data Mining to Electronic Commerce. Springer US, pp.115-153, 2001.
[4]Langheinrich M., Nakamura A., Abe N., Kamba T., and Koseki Y., “Unintrusive customization techniques for Web advertising,” Computer Networks 31(11), pp. 1259-1272, 1999.
[5]Y. Jang , T. Lee, K. Kim, and W. Lee, “Keyword Management System based on Ontology for Contextual Advertising,” 6th Internationa lConference on Advanced LanguageProcessing and Web Information Technology,pp. 440-445, 2007.
[6]H. C. Huang, M. S. Lin, and H. H. Chen, “Analysis of Intention in Dialogues Using Category Trees and Its Application to Advertisement Recommendation,” 3rd International Joint Conference on NaturalLanguage Processing, pp. 625-630, 2008.
[7]G. Salton and C. Buckley, “Term-weighting approaches in automatic text retrieval,” Information processing & management, Vol. 24, pp. 513-523, 1988.
[8]Chen P. I., and Lin S. J., “Automatic keyword prediction using Google similarity distance,” Expert Systems with Applications, 37(3), pp. 1928-1938, 2010.
[9]W. Y. Ma and K. J. Chen, “Design of CKIP Chinese word segmentation system,” Chineseand Oriental Languages Information Processing Society, Vol. 14. No. 3. pp. 235-249, 2005.
[10]Chinese Knowledge Information Processing Group(CKIP), http://ckipsvr.iis.sinica.edu.tw/
[11]YAHOO Keyword categories, Available on line (Retrived: May 2015) https://tw.ysm.emarketing.yahoo.com/soeasy/
[12]R. L. Cilibrasi and P. M. Vitanyi, “The Google similarity distance,” IEEE Transactions on Knowledge and Data Engineering, Vol. 19, No. 3, pp. 370-383,2007.
[13]M. Shatnawi and N. Mohamed, “Statistical Techniques for Online Personalized Advertising: A survey,” 27th Annual ACM Symposium onApplied Computing, pp. 680–687, New York, 2012.
[14]R. Studer, V.R. Benjamins, and D. Fensel, “Knowledge engeering: principles and methods,” Data and knowledge engineering, Vo. 25, No. 1, pp. 161-197, 1998.
[15]T. R. Gruber, “A translation approach to portable ontology specifications, Knowledge Acquisition,” Vol. 5, No. 2, pp. 199-200, 1993.
[16]M. Uschold, and M. Gruninger, “Ontologies: principles, methods and applications,” Knowledge Engineering Review, Vo. 11, No. 2, pp. 93-136, 1996.
[17]J. Li, and O.R. Zaiane, “Combining usage, content, and structure data to improve Web site recommendation,” Proc. of the 5th Int. Conf. on Electronic Commerce and Web Technologies (EC-Web), 2004.
[18]C. Shahabi, F. Banaei-Kashani, Y. Chen, and D. McLeod, “Yoda: An Accurate and Scalable Web-based Recommendation System,” Proc. of COOPIS., 2001.
[19]J. S. Hsu and T. C. Hung, “The Use of Ontology in Case-based Reasoning System for Itinerary Recommendation,” Journal of Information Management (in Chinese),Vol. 9, pp.31-61,2004
[20]R.-D. Ou,“Using Ontology and Universal User Profile for Personalized Recommendation”, Master Thesis (in Chinese), Department of Information Management, Chaoyang University of Technology, 2006.
[21]D. W. Hosmer Jr, S. Lemeshow, and R. X. Sturdivant, Applied Logistic Regression, Vol. 398,John Wiley & Sons,2006.
[22]Berkson J., “A statistically precise and relatively simple method of estimating the bio-assay with quantal response, based on the logistic function,” Journal of the American Statistical Association, 48(263), pp. 565-599, 1953.