Decision Support And Profit Prediction For Online Auction


Published on

Online auction has become a very popular e-commerce trans-action type. The immense business opportunities attract a lot of individuals as well as online stores. With more sellers engaged in, the competition between sellers is more intense. For sellers, how to maximize their pro¯t by proper auction setting becomes the critical success factor in online auction market. In this paper, we provide a selling recommendation service which can predict the expected pro¯t before listing and, based on the expected pro¯t, recommend the seller whether to use current auction setting or not. We collect data from ¯ve kinds of digital camera from eBay and ap-
ply machine learning algorithm to predict sold probability and end-price. In order to get genuine sold probability and end-price prediction (even for unsold items), we apply probability calibration and sample selection bias correction when building the prediction models. To decide whether to list a commodity or not, we apply cost-sensitive analysis to decide whether to use current auction setting. We compare the pro¯ts using three di®erent approaches: probability-based, end-price based, and our expected-pro¯t based recommendation service. The experiment result shows that our recommendation service based on expected pro¯t gives higher earnings and probability is a key factor that maintains the pro¯t gain when ultra cost incurs for unsold items due to

Published in: Business, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • Support Vector Machine(SVM) 是近年來相當流行的分類演算法,並普遍使用於解決 各種分類問題。雖然 SVM 並不直接提供預測的機率值, 不過我們能從資料點至平面 的距離來衡量預測該類別的信心度,並將此信心度轉換正規化至 0~1 區間來代表真實機 率值,但是信心度與真實機率並不一定存在著線性的關係,所以我們無法透過簡單的正 規化公式來進行轉換。
  • Decision Support And Profit Prediction For Online Auction

    1. 1. Authors / Chia-Hui Chang and Jun-Hong Lin Presenter / Meng-Lun Wu Decision Support and Profit Prediction for Online Auction Sellers
    2. 2. Outline <ul><li>INTRODUCTION </li></ul><ul><li>DATA </li></ul><ul><li>TO SELL OR NOT TO SELL? </li></ul><ul><li>END-PRICE PREDICTION </li></ul><ul><li>SOLD PROBABILITY CALIBRATION </li></ul><ul><li>EXPRIMENTS </li></ul><ul><li>CONCLUSIONS </li></ul>
    3. 3. INTRODUCTION <ul><li>Online auction has become a very popular e-commerce transaction type. </li></ul><ul><li>The main issue is to find an auction setting that could maximize sellers profit is a challenging problem. </li></ul><ul><ul><li>profits prediction. </li></ul></ul><ul><li>Ghani and Simmons apply three models for end-price prediction. </li></ul><ul><ul><li>regression, multi-class classification and multiple binary classification. </li></ul></ul>
    4. 4. INTRODUCTION (cont.) <ul><li>Some researchers suggest the use of predicted end-price for selling strategy, but they didn’t address how to apply predicted end-price in decision making. </li></ul><ul><li>This paper proposed the use of cost-sensitive decision making to resolve whether to list a commodity or not. </li></ul><ul><li>Authors use profit gain over average profit of similar items sold by similar sellers. </li></ul>
    5. 5. DATA <ul><li>Authors collect the digital cameras auction data from eBays, and choose the period of two months in March-April 2006. </li></ul><ul><li>These data are classified into 3 classes: </li></ul><ul><ul><li>Seller Features, Items Features, Auction Features </li></ul></ul>
    6. 6. DATA (cont.)
    7. 7. DATA (cont.)
    8. 8. Two tasks
    9. 9. TO SELL OR NOT TO SELL? <ul><li>This paper have three approaches of decision criteria: </li></ul><ul><li>Sold probability Based Approach </li></ul><ul><ul><li>When the predicted sold probability is greater than 50%. </li></ul></ul><ul><ul><ul><li>P(c=1|x) > P(c=-1|x) </li></ul></ul></ul><ul><li>End-price Based Approach </li></ul><ul><ul><li>When the predicted end-price is higher than the average end-price of similar items </li></ul></ul><ul><ul><ul><li>y(x) > Avgy(x) </li></ul></ul></ul>
    10. 10. TO SELL OR NOT TO SELL? (cont.) <ul><li>Expected profit Based Approach </li></ul><ul><ul><li>When the expected profit gain for suggesting sell is greater than zero. </li></ul></ul><ul><ul><ul><ul><li>P(c=1|x)[Profit(x) – AvgP(x)]>P(c=-1|x)[lc(x)+uc] </li></ul></ul></ul></ul>
    11. 11. END –PRICE PREDICTION <ul><li>There are two types of auction end-price predictions </li></ul><ul><ul><li>Static : Multiple Binary Classification </li></ul></ul><ul><ul><li>Dynamic : Sample Selection Bias Correction </li></ul></ul><ul><li>Ghani and Simmons proposed Multiple Binary Classification prediction method with neural network algorithm get the better accuracy. </li></ul><ul><li>Authors use multiple binary classifiers with two-class estimation to predict the end-price for auction listing items. </li></ul>
    12. 12. TWO–CLASS ESTIMATION 47.5
    13. 13. Sample Selection Bias Correction <ul><li>Sample selection bias problem has been studied in econometrics and statistics. </li></ul><ul><li>There are four kinds of sample selection bias: </li></ul><ul><ul><li>Complete Independent </li></ul></ul><ul><ul><li>Feature Bias </li></ul></ul><ul><ul><li>Class Bias </li></ul></ul><ul><ul><li>Complete Bias </li></ul></ul>
    14. 14. SOLD PROBABILITY CALIBATION <ul><li>To estimation the sold probability for an item, we can use any binary classification to train a model for predicting whether an item would be sold (c=1) or not (c=-1). </li></ul><ul><li>In this paper, we use SVM as our classifier. </li></ul><ul><li>SVM produces an uncalibrated value that is not a probability. </li></ul><ul><li>To yield well calibrated probability, we adopt Platt Scaling to transfer a value f(x i ) into probability P(c=1|x i ). </li></ul>
    15. 15. EXPERIMENTS <ul><li>70% data as training data and 30% as testing data. </li></ul><ul><li>The experiments divided into three parts: </li></ul><ul><ul><li>The performance on item sold prediction </li></ul></ul><ul><ul><li>The accuracy of end-price prediction </li></ul></ul><ul><ul><li>Compares three performance of three approaches </li></ul></ul><ul><ul><ul><li>Probability-based </li></ul></ul></ul><ul><ul><ul><li>End-price based </li></ul></ul></ul><ul><ul><ul><li>Expected profit based </li></ul></ul></ul>
    16. 16. Accuracy of Sold Probability <ul><li>Authors use SVM as our classifier for predicting. </li></ul><ul><li>The accuracy of prediction whether an item will be sold I about 93.5% and 76.2% for auction listing and BuyItNow listing. </li></ul>
    17. 17.
    18. 18. END-price Prediction <ul><li>To use multiple binary classifiers for end-price prediction, we need to decide the reference value for each binary classifier. </li></ul><ul><li>Authors exclude the highest and lowest end-prices, and use 10% window of the average end-price as the interval size. </li></ul>
    19. 19.
    20. 20.
    21. 21. Profit Gain Comparison <ul><li>In order to validate that end-price alone does not guarantee high profit due to low sold probability, authors use profit gain over other sellers to compare the three approaches. </li></ul><ul><li>Sell all approach as the baseline for comparison. </li></ul>
    22. 22. Auction Listing
    23. 23. BuyItNow Listing
    24. 24. CONCLUSIONS <ul><li>A critical question for online auction sellers is to find an auction setting. </li></ul><ul><li>Authors apply machine learning algorithms for end-price prediction and sold probability estimation. </li></ul><ul><li>For end-price prediction, since we can only use old items to train the classifiers, the model could under-estimate the end-price for unsold items. </li></ul>
    25. 25. CONCLUSIONS (cont.) <ul><li>As shown in the experiments, approaches that don’t involve the consideration of sold probability have deteriorating profit as ultra cost increase. </li></ul><ul><li>This proves our argument that sold probability should also be considered for profit maximization. </li></ul><ul><li>The approach based on both probability and end-price have the best performances among all other approaches. </li></ul>