A Classification Approach for Movie Recommender System 指導教授:黃三益 老師 學生: M 黃于珊 M 李界寬 M 程尚文
Agenda Introduction Motivation and background Determination of data set The Data Mining Procedure Conclusion and Limitation
1.MOTIVATION AND BACKGROUND 2.DETERMINATION OF DATA SET INTRODUCTION
Motivation and background Dataset 來源自 GroupLens ◦ (Research lab in the Department of Computer Science and Engineering at the University of Minnesota ; 線上電影推薦系統 -MovieLensMovieLens ( ) ◦ 加入會員,評價隨機選出的數部電影,即 可享受到網站給予的五部電影之推薦,並 附上預測使用者喜好該電影的程度。 We all loves movies Find the rule
Determination of data set 使用 MovieLens 目前提供兩種 Datasets 的其中一種。 ◦ 內容包含 1682 部電影, 943 使用者,共 100,000 ratings 。 ◦ 提供足夠的樣本規模,讓我們可以適當的 建立和測試模型。
1.DATA MINING PROCEDURE:10 STEP 2. CONCLUSION AND LIMITATION The Data Mining Procedure
Step 1. Translate the business problem into a data mining problem 電影種類與數目相當繁多,如何在眾多 的電影中可以快速的找到符合自己偏好 的電影 ? ◦ 電影推薦系統 ◦ 縮短搜尋時間 ◦ Find the Rule 年齡、職業、性別等之偏好那些種類的電影 ◦ Potential customers
Step 2. Select appropriate data 線上電影推薦系統 -MovieLens Research lab in the Department of Computer Science and Engineering at the University of Minnesota ; 資料來源自加入其網站的會員對電影所作的評價與 會員的相關個人資料 其所提供的 Dataset 內容包含 1682 部電影, 943 使 用者,共 100,000 ratings 。
Step 3. Get to know the data(1/2) This data has been cleaned up ◦ users who had less than 20 ratings ◦ did not have complete demographic information
Step 3. Get to know the data(2/2) Attribute nameDescriptionDomain Age User 年齡 1: “Under 18” , 18: "18-24“ 25: “25-34” , 35: "35-44" 45: “45-49” , 50: "50-55“ 56: "56+” Gender User 性別 "M" 代表男性, "F" 代表女性 Occupation User 職業 0: "other" or not specified 1: “academic/educator” 2: "artist" 3: “clerical/admin” 4: "college/grad student“ And so on…… Movie Kind 電影類型 * Action * Adventure * Animation * Children‘s * Comedy * Crime * Documentary * Drama * Fantasy * Film-Noir * Horror * Musical * Mystery * Romance * Sci-Fi * Thriller * War * Western
Step 4. Create a model set Data Source – MovieLens (The GroupLens Research Project at the University of Minnesota) Data Characteristics: – 100,000 ratings (1-5) from 943 users on 1682 movies – Each user has rated at least 20 movies – seven-month period from September 19th, 1997 through April 22nd, 1998 – With complete demographic information
Step 5. Fix problems with the data Variable with too many values ◦ Movie kind ◦ Occupation ◦ We do not consider variables such as ZipCode and rate
Step 6.Transform data to bring information to the surface We skip this step due to the uselessness of transforming data into different formats
Step 7. Build models Data mining tool: ◦ Weka Explorer Classifier ◦ Decision tree methods ◦ using C4.5 algorithm Performs well on both accuracy and speed
Weka: the software
Step8. Assess Model Confusion Matrix Table 1. Confusion Matrix of Classifier C4.5 from Training Set The Kind of MovieRomanceThrillerWar Romance2,5767,46538 Thriller1,74215,64353 War1,0956,42890
Step8. Assess Model Detailed Accuracy Table 2. Detailed Accuracy of Classifier C4.5 from Training Set ClassTP RateFP RatePrecisionRecallF-Measure Romance Thriller War
Step8. Assess Model Other Information Table 3. The Results of Classifier C4.5 from Training Set Correctly Classified Instances 18,309Rate : % Incorrectly Classified Instances 16,821Rate : % Kappa statistic Mean absolute error Root mean squared error Relative absolute error % Root relative squared error % Total Number of Instances 35,130
Step 8. Assess Model Decision Tree ◦ Number of Leaves : 118 ◦ Size of the tree : 216
Step 9. Deploy Model It’s difficult to deploy, because ◦ Computer’s resources are not enough ◦ Difficult to implementation
Conclusion and Limitation Classification Approach : C4.5 → Decision Tree Data Set : 35,130 data Limitation ◦ Hardware and software don’t support enough to mining more data to find more interest and complete rules.
Thanks For Your Attention.