Personalization of Collaborative Fileting by Item Metadata User Profiling / Osama Abdelhameed Ahmed Yassin Haggag

Supervisor: Samhaa El-Beltagy

Thesis (M.A.)—Nile University, Egypt, 2014 .

"Includes bibliographical references"

Contents:
Chapter 1 Introduction ........................................................................................................ 7
1.1 Motivation ........................................................................................................... 7
1.2 The Importance of Recommendation Research .................................................. 8
1.3 Problem Statement ............................................................................................. 9
1.4 Outline of the Thesis ......................................................................................... 10
Chapter 2 Recommender Systems Overview .................................................................... 11
2.1 Taxonomy of Recommendation Systems .......................................................... 11
2.2 Recommendation Techniques and Algorithms ................................................. 14
2.2.1 Memory-based Algorithms ........................................................................ 14
2.2.2 Model-based Algorithms ........................................................................... 17
2.2.3 Hybrid Algorithms ...................................................................................... 18
2.2.4 Similarity .................................................................................................... 18
2.3 Challenges of the Recommendation Paradigm ................................................. 20
Chapter 3 Recommendation Environment of the PlayStation Network ........................... 22
3.1 Characteristics of the System ............................................................................ 22
3.1.1 Algorithms of the System .......................................................................... 22
3.1.2 Time Intervals of Recommendation .......................................................... 23
3.1.3 Down-weighting/Forgetting Old Items ...................................................... 24
3.1.4 Parallelization with OpenMP through Matlab MEX .................................. 24
3.2 Generating Recommendations .......................................................................... 25
3.3 Blending of Algorithms ...................................................................................... 25
Chapter 4 Evaluation Metrics for Recommender Systems ............................................... 27
4.1 Error Rate (errRate@X) ..................................................................................... 27
4.2 R-Precision (R-PRC) ............................................................................................ 29
4.3 Reverse Mean Reciprocal Ranking (RMRR) ....................................................... 30
4.4 Top 100 hits ....................................................................................................... 32
Chapter 5 Description and Analysis of the Used Dataset ................................................. 33
5.1 Overview ............................................................................................................ 33
5.2 Identifying Item Categories ............................................................................... 35
5.3 Segmenting the Dataset .................................................................................... 37
5
5.4 Analysis of User Behavior .................................................................................. 41
5.4.1 Number of views of an item before a purchase: ....................................... 41
5.4.2 Time between first add-to-cart and purchase of an item ......................... 43
5.4.3 Purchase Statistics ..................................................................................... 44
5.5 Training and Test Splits ...................................................................................... 46
Chapter 6 Configuration of the Simulation Environment and Approximations ................ 47
6.1 User Evaluation Criteria ..................................................................................... 47
6.2 Approximations ................................................................................................. 48
6.2.1 Similarity Matrix ........................................................................................ 48
6.2.2 User Activity............................................................................................... 48
6.3 Conclusion ......................................................................................................... 51
Chapter 7 Experimental Approaches ................................................................................. 52
7.1 Cold-start Effect Analysis ................................................................................... 53
7.2 Category profiling .............................................................................................. 54
7.3 Price profiling..................................................................................................... 56
Chapter 8 Experimental Results ........................................................................................ 58
8.1 Sandbox Setup ................................................................................................... 58
8.2 Category Reranking ........................................................................................... 59
8.2.1 ErrRate@3 ................................................................................................. 60
8.2.2 R-PRC ......................................................................................................... 61
8.2.3 Top100 ....................................................................................................... 63
8.2.4 RMRR ......................................................................................................... 64
8.2.5 Interpreting the results ............................................................................. 66
8.3 Price Reranking .................................................................................................. 66
8.3.1 ErrRate@3 ................................................................................................. 66
8.3.2 R-PRC ......................................................................................................... 68
8.3.3 Top100 ....................................................................................................... 69
8.3.4 RMRR ......................................................................................................... 71
8.3.5 Interpreting the results ............................................................................. 72
8.4 Mixing Category and Price Profiling .................................................................. 73
8.4.1 ErrRate@3 ................................................................................................. 73
8.4.2 R-PRC ......................................................................................................... 74
8.4.3 Top100 ....................................................................................................... 76
6
8.4.4 RMRR ......................................................................................................... 77
8.4.5 Interpreting the results ............................................................................. 79
8.5 Cold-start Effect ................................................................................................. 79
Chapter 9 Conclusion, Drawbacks and Future Work......................................................... 82
9.1 Conclusion ......................................................................................................... 82
9.2 Drawbacks ......................................................................................................... 83
9.3 Future Work ...................................................................................................... 83
References .........................................................................................................................

Abstract:
Recommendation systems typically deal with enormous amounts of data making it difficult to implement complex solutions to the recommendation problem. The scalability constraints of the algorithms used is often the issue. In this work, a simple and direct method for the improvement of the recommendation performance is investigated. An items-frequently-bought-together collaborative filtering algorithm is augmented with information about user activity in the Playstation store. This work uses information of user item views, add-to-cart, and purchase activities to calculate user profiles. This profile contains the user preferences for item categories and item prices. It is then used to reorder collaborative filtering recommendations to reflect the collected user profile. In addition, the effect of using all available user activities to lessen the cold-start effect is investigated. Results show improvement in the performance of the collaborative filtering algorithm when augmented with extra information, more so on data of the same type, e.g. films. Results also show that the cold-start effect can be lessened using view and add-to-cart information in addition to purchase information.

Text in English, abstracts in English.