Browsing by Author "Shih, Wen-Yueh"

Now showing 1 - 7 of 7

A Classification-Based Elephant Flow Detection Method using Application Round on SDN Environments
(2017-01-01)
Huang, Yuan-Hao
;
Shih, Wen-Yueh
;
Huang, Jiun-Long
;
資訊工程學系
;
Department of Computer Science
We propose in this paper a classification-based elephant flow detection method for SDN environments. The proposed elephant flow detection method consists of two classifiers running on the switch and the controller, respectively. For better performance, the concept of application round is used to build the classifier running on the controller. Experimental results show that our elephant flow detection method is able to obtain high recall and F-measure.
A Gamma-based Regression for Winning Price Estimation in Real-Time Bidding Advertising
(2017-01-01)
Zhu, Wen-Yuan
;
Shih, Wen-Yueh
;
Lee, Ying-Hsuan
;
Peng, Wen-Chih
;
Huang, Jiun-Long
In Real-Time Bidding (RTB) advertising, estimating the winning price is an important task in evaluating the bid cost of bid requests in Demand-Side Platforms (DSPs). The prior works utilize censored linear regression for winning price estimation by considering both winning and losing bid records. In the traditional regression models, the winning price of each bid request is based on Gaussian distribution. However, the property of Gaussian distribution is not suitable for the winning price of each bid request, and it is hard to link the physical meaning of Gaussian distribution and the winning price. Therefore, in this paper, based on our observation and analysis, the winning price of each bid request is modeled by a unique gamma distribution with respect to its features. Then we propose a gamma-based censored linear regression with regularization for winning price estimation. To derive the parameters of our proposed complicated model based on bid records, our approach is to divide this hard problem into two sub-problems, which are easier to solve. In practice, we also provide four heuristic initial parameter settings that are able to greatly reduce the computation cost when deriving the parameters. The experimental results demonstrate that our approach is highly effective for estimating the winning price compared with the state-of-the-art approaches in three real datasets.
An Expected Win Rate-Based Real Time Bidding Strategy for Branding Campaign by the Model-Free Reinforcement Learning Model
(2020-01-01)
Shih, Wen-Yueh
;
Lu, Yi-Shu
;
Tsai, Hsiao-Ping
;
Huang, Jiun-Long
;
資訊工程學系
The bidding strategy plays the most important role to help the Demand Side Platforms (DSPs) making bidding decisions on a large number of bid requests in Real Time Bidding (RTB) to satisfy the different objectives of campaigns under the lifetime and budget constraints. In this paper, we focus on branding campaign whose objective is to obtain as many impressions as possible under the lifetime and budget constraints. To achieve the objectives of branding campaigns, we propose a novel expected win rate-based bidding strategy for branding campaign under the lifetime and budget constraints by utilizing a model-free reinforcement learning model. Specifically, to prevent missing good opportunities resulting from submitting extremely low bid prices, the concept of the base winning price is introduced to determine the lower bound of expected winning price. In addition, to obtain more impressions, the concept of the DSP-specified budget spending plan is proposed to determine the proper winning prices. The base expected win rate is then calculated based on the base winning price and the winning price determined by the DSP-specified budget spending plan. Since RTB is a dynamic environment, we propose a novel expected win rate-based bidding strategy named EWDQN which utilizes Deep Q Network (DQN) to dynamically determine the expected win rate according to the base expected win rate and the current status of the RTB market, and then determines the bid price according to the expected win rate. To the best of our knowledge, this is the first research applying the reinforcement learning technique on the bidding strategies for branding campaign. To measure the performance of EWDQN, several experiments are conducted on two real datasets. Experimental results show that EWDQN outperforms the-state-of-the-art bidding strategies for branding campaign in terms of the number of obtained impressions and CPM (cost per thousand impressions).
An expected win rate-based real-time bidding strategy for branding campaigns on display advertising
(2019-12-01)
Shih, Wen-Yueh
;
Huang, Jiun-Long
;
資訊工程學系
;
Department of Computer Science
For branding campaigns, the demand-side platforms (DSPs) in real-time bidding (RTB) usually need to win as many impressions as possible to inform most audiences about the product messages under constraints on budgets, campaign lifetimes and budget spending plans. In this paper, we propose a novel bidding strategy by introducing the concept of expected win rate. With the proposed expected win rate-based bidding strategy, the DSP can dynamically adjust the expected win rate for each incoming bid request based on factors such as the predicted number of bid requests in the near future, the remaining budget and the remaining lifetime of the campaign. The experimental results show that the proposed bidding strategy has a lower cost per thousand impressions and cost per clicks than existing pacing model-based bidding strategies for branding campaigns with the same budgets and budget spending plans.
On successive point-of-interest recommendation
(2019-05-01)
Lu, Yi-Shu
;
Shih, Wen-Yueh
;
Gau, Hung-Yi
;
Chung, Kuan-Chieh
;
Huang, Jiun-Long
With the increasing popularity of location-based social networks (LBSNs), users are able to share the Point-of-Interests (POIs) they visited by check-ins. By analyzing the users' historical check-in records, POI recommendation can help users get better visiting experience by recommending POIs which users may be interested in. Although recent successive POI recommendation methods consider geographical influence by measuring the distances among POIs, most of them ignore the influence of the regions where the POIs are located. Therefore, we propose in this paper two models to tackle the problem of successive POI recommendation. First, a feature-based successive POI recommendation method, named UGSE-LR, is proposed to take the influence of regions, named regional influence, into consideration when recommending POIs. UGSE-LR first splits an area into grids for estimating regional influence. Then, UGSE-LR applies Edge-weighted Personalized PageRank (EdgePPR) for modeling the successive transitions among POIs. Finally, UGSE-LR fuses user preference, regional influence and successive transition influence into a unified recommendation framework. In addition, with the aid of Recurrent Neural Network (RNN), we propose a latent-factor based successive POI recommendation method, named PEU-RNN, to integrate the sequential visits of POIs and user preference to recommend POIs. First, PEU-RNN adopts the word embedding technique to transform each POI into a latent vector. Then, RNN is used to recommend the POIs depend on the users' historical check-in records. Experimental results on two real LBSN datasets show that our methods are more accurate than the state-of-the-art successive POI recommendation methods in terms of precision and recall. In addition, experimental results also show that PEU-RNN is suitable for the datasets with many check-in records, while UGSE-LR is suitable for the datasets with moderate check-in records.
Predicting Traffic of Online Advertising in Real-time Bidding Systems from Perspective of Demand-Side Platforms
(2016-01-01)
Lai, Hsu-Chao
;
Shih, Wen-Yueh
;
Huang, Jiun-Long
;
Chen, Yi-Cheng
;
資訊工程學系
Online advertising has been all the rage these years. Budget control and traffic prediction turn out to be important issues for the demand-side platforms (DSPs). However, DSPs cannot easily grab the information of audiences and media platforms. Although DSPs might have the information immediately, it is still hard to response the request of advertisements in real-time due to the high volume of features. Therefore, we propose a method predicting traffic of requests from perspective of DSPs. The features we used are simple to be extracted from historical data. The prediction model we chose is regression model with closed-form solution. Both the features and regression model make our prediction adaptive in real-time systems. Our method can detect traffic anomalies and prevent it from overwhelming prediction. Moreover, our method can also keep pace of the trend. Experiment results show that our method's error rate of prediction is about 0.9% in total, and 10% per time unit.
SEM: A Softmax-based Ensemble Model for CTR Estimation in Real-Time Bidding Advertising
(2017-01-01)
Zhu, Wen-Yuan
;
Wang, Chun-Hao
;
Shih, Wen-Yueh
;
Peng, Wen-Chih
;
Huang, Jiun-Long
In Real-Time Bidding (RTB) advertising, evaluating the Click-Through Rate (CTR) of a bid request and an ad is important for bidding strategy optimization on Demand-Side Platforms (DSPs). The regression-based approaches are popular for CTR estimation in RTB since this kind of approach is highly efficient and scalable. The information of the bid request and the ad contains categorical attributes (such URL) and numerical attributes (such ad size). To vectorize the information for the input of regression-based approaches, the categorical attributes will be expanded to several binary features in general. However, some categorical attributes have infinite possible values (such as URL). Thus, for these attributes, only observed values in training will be transformed into binary features. If there is a new attribute or value in online environment, this information will be lost after vectorization. In this paper, we first exploit the feature hashing trick to transform the categorical and numerical attributes into the large fixed size vector. Since the vector is large and sparse, we propose a Softmax-based Ensemble Model, SEM, which adopts only a few key features after feature hashing for CTR estimation. The experimental results demonstrate that our proposed approach is able to adapt to the harsh environments in RTB, and outperforms the state-of-the-art approaches effectively when only less than 50 features are adopted in two real datasets.