Prediction of Hits in E commerce based on Exponential Smoothing Technique

2017 2nd International Conference on Advances in Management Engineering and Information Technology (AMEIT 2017) ISBN: 978-1-60595-457-8

Prediction of Hits in E-commerce based on Exponential

Smoothing Technique

Jin LIAN

and Li LI

Sichuan TOP IT Vocational Institute, Chengducity,Sichuan province,China

Keywords: Hits, Exponential smoothing, Optimal smoothing constant, Prediction.

Abstract. The prediction of views on E-commerce businesses is a great thing, it directly affects the

trading volume of goods. Through the research of exponential smoothing, this article determines the optimum smoothing constant, establishes Second Exponential Smoothing model over a period of time, fits the views trend using the model, and predicts the recent hits according to the model, so as to make reference for businessmen.

Introduction

In recent years, E-commerce develops rapidly involving more and more areas, the transaction entity keeps increasing, and the size of transaction scales is growing. According to the data about the first half of 2015 provided by China E-business Research Center, E-commerce transactions has reached 7.63 trillion, B2B transactions accounts the largest reaching 5.8 trillion, and C2C transactions is 1.61 trillion up 48.7% from previous year. The daily business transaction has exceeded 100 billion according to data provided by Tmall about Double 11 event in 2016.

Although the general trend in E-commerce has been good development prospects, E-commerce businessmen pay more attention to their online stores turnover. Online stores turnover is directly related to online traffic, and the hit is an important indicator of online traffic.

Hits in E-commerce is also known as online store views, is the number of users clicking on the web when they access the online store. Users click on each page distributed in the store, some click in the commodity page, some online homepage and category pages, and each click can clearly recorded by hot map. Through each location click on statistics, we can analyze popularity of store each page and each location, so as to better understand the decoration and marketing problem of the store, and then contribute to more deals through improvement, so the hits directly affect the transaction of online stores, it is crucial for online merchants to predict reasonably hits through each part of pages. Through study of Exponential Smoothing Technique, the article reasonably predict hits, thereby provide references for businessmen.

The Basic Theory of Exponential Smoothing Technique

The Generation of Exponential Smoothing Technique

Exponential Smoothing Technique is a forecasting technique based on time series, is one of the most commonly prediction methods, put forward by famous American economist and mathematicians Brown [1]. Brown holds that the stability and regularity of time series can be conveniently postponed, and people can eliminate the impact of random factors based on computing time series data of smoothing, so can predict economic changes [3].The character of Exponential Smoothing Technique is that the latest time data (observations) is given the biggest weight, other prediction data are given weight from large to small, forecast data can not only reflect the latest information, but also reflect the information before, so as to make prediction results close more to the actual[2].

The Type of Exponential Smoothing Technique

according to the prediction model. Depending on the number of smoothing, Exponential Smoothing Technique can be divided into three techniques: First Exponential Smoothing Technique, Second Exponential Smoothing Technique and Third Exponential Smoothing Technique, etc.

First Exponential Smoothing Technique.

First Exponential Smoothing Technique is applicable to forecast stationary data, and can predict well the next issue of data values when time series have no obvious changing tendency. Set the time sequencesy₁, y₂,…, y_t,...,then the exponential smoothing formula is

(1) (1)

1

(1 ) , 0 1, 3

t t t

S =ay + −a S− <a≤ a≥ ₍₁₎

St(1) above is an First Exponential Smoothing value in the t period; a is the smoothing constant, 0<a≤l,t≥3;yt is the actual data values in the t period. After the above formula is expanded,

then leveled off exponential smoothing model is:

1

ˆ_{t t} (1 )ˆ_t

y+ =ay + −a y ₍₂₎

That is to say, the first exponential smoothing value in the t period will be the predictive value of the t+1 period.

Second Exponential Smoothing Technique. When time sequence change is not stable but

linear, the use of Exponential Smoothing Technique will produce obvious lag error, then Exponential Smoothing Technique is no longer applicable, and need to be revised. The modified method is to make exponential smoothing again on the basis of the First Exponential Smoothing Technology, so as to find out the regularity of lag deviation and build a prediction model.

If the First Exponential Smoothing value is (1) t

S , the Second Exponential Smoothing value is (2)

t

S , then the calculation formula is

(2) (1) ( 2)

1 (1 )

t t t

S =aS + −a S− ₍₃₎

Expanding the formula, the Second Exponential Smoothing Model which tends to be linear is

ˆ_{t T t t}

y+ =a +b T ₍₄₎

Among the Model, a_tis the intercept of linear equation, b_tis the slope: ₂ (1) ( 2)

t t t

a = S −S ,

(1) ( 2)

( )

1

t t t

a

b S S

a

= −

− ; t is the current period number; And T is the period number from the current

period to predictive period; yˆ_{t T}+ is the predicted value for the t + T period.

Third Exponential Smoothing Technique. When the time sequence change is nonlinear, Third

Exponential Smoothing Technique is useful. The method is to make exponential smoothing again on the basis of Second Exponential Smoothing Technology. The formula is

(3) ( 2) (3)

1 (1 )

t t t

S =aS + −a S− ₍₅₎

Expanding the formula, Third Exponential Smoothing model tending to be quadratic curve becomes

2 t T t t t

y+ =a +b T+c T ₍₆₎

Among the formula, ₃ (1) ₃ (2) (3)

t t t t

a = S − S +S ， (1) (2) (3)

2[(6 5 ) 2(5 4 ) (4 3 ) ] 2(1 )

t t t t

a

b a S a S a S

a

= − − − + −

− ，

2

(1) (2) (3)

2[ 2 ]

2(1 ) t

a

c St St St

a

= − +

Solving Steps of Exponential Smoothing Technique

(1) Collect data information and process. Because time series can conveniently extend to the future

according to the recent past situation, so as far as possible collect data information which is close to

the prediction time, in order to better forecast.

(2) Make a chart according to the processed data, analyze the data change tendency, choose reasonable forecast model. Choose First Exponential Smoothing model when data changes more smoothly; Choose Second Exponential Smoothing model when data change to be linear; Choose Third Exponential Smoothing model when data change to be a quadratic curve type.

(3) Determine the smooth index. Choosing appropriate smooth value a often directly affect that the forecast is successful or not. Value a _{is often determined by experience, Bower-man and} O'Conner recommended that it is values better to control the range between 0.1 ~ 0.3 [1]. But usually determine the smoothing constant according to the change of time series, a _{should be took smaller} if the time series changes more smoothly, such as 0.05 ~ 0.2; The value a _{should be a little bigger} such as 0.1 ~ 0.4 when time series have little change on the whole; When time series change a lot, and the overall change has obvious and rapidly increasing or decreasing, the value a should be choose larger smoothing coefficient, such as 0.6 to 1. [7] In a practical application, the value a _is often determined basing on experience and the actual situation, and the value a _{with the least} prediction error will be choose through calculating and comparing multiple values ofa.

(4) According to the established prediction model, put parameters among the model, and then get the predicted value. [5]

Prediction of Online Page Hits Based on Exponential Smoothing Technique

The author collects hits of customer for fixed commodity pages through monitoring online store, and draws a curve graph from Feb.10th, 2016 to Feb.17th, 2016, as shown in figure 1.

Figure 1. Hits of goods from Feb.10th to Feb.17th.

From the figure 1 , the hits of goods tend to general linear growth from Feb.10th to Feb.17th, and hits on Feb.17th reach the peak 509. So according to Exponential Smoothing Technique, clicks on Feb.18 should be predicted adopting Second Smoothing model

Establishing First Exponential Smoothing Forecasting Model

From Formula (2) , the First Smoothing Forecasting model is: yˆ_t+₁=ay_t+(1−a y)ˆ_t, in which 0<a≤

l,t≥3. At this point, it is necessary to determine the smoothing constant a.

precision of RMSE is higher than MSE, so the article determines the smoothing constant a by use of RMSE. According to the established a_{, after the trial calculation with established} a, it is

conclusion that the smoothing constant a _{0.7 can reach standard minimum error, as shown in} Table 1.

Table 1. Smoothing Constant Spread trial calculating sheets.

Establishing Second Exponential Smoothing Forecasting Model

Upon data of First Smoothing Model, Second Exponential Smoothing model can be made through smoothing again. According to Second Exponential Smoothing model: yˆ_{t T}+ =a_t+b T_t ,the intercept

t

a and slope b_t can be calculated.

(1) ( 2) 2

t t t

a = S −S =455.141

(1) ( 2)

( )

1

t t t

a

b S S

a

= −

− =15.2201

So yˆ_{t T}+ =a_t+b T_t =455.141+15.2201*1=470.3611，the predicted value is 470 in round numbers, and the results are shown in Figure 2.

Figure 2. Goods Hits Forecast Figure.

Determining the Best Prediction

Although better predict hits can be reached by Second Smoothing, but the value a _{is upon the} experience, there are a lot of subjectivity, and the smoothing constant a _{is uncertain to be the} best. Therefore, the key is to determine the optimal smoothing constant while predicting. In the article, the optimal smoothing constant a _{is 0.7811, using planning method with evaluation of}

RMSE, the standard error is 18.982. It is obvious that this standard error is less than the standard error of 19.75019 upon experience, so the best smoothing constant value a _{is more reasonable.}

Then use the best smoothing constant a _{for First Exponential Smoothing and Second} Exponential Smoothing, finally make calculation using Second Exponential Smoothing forecasting

model: ₂ (1) ( 2)

t t t

a = S −S _{=472.75, (} (1) ( 2)₎ 1

t t t

a

b S S

a

= −

− =97.378, yˆt T+ at b Tt

= +

Figure 3. The Best Predictive Value.

Conclusion

Hits of the store goods directly affect the product conversion rate, and is important for the commodity search optimization, keyword promoting etc. In this article, hits of customers for fixed commodity pages is collected and processed in eight days from Feb.10th to Feb.17th. study of clicks based on Exponential Smoothing Technique and the establishment of the Secondary Exponential Smoothing model, page clicks on the actual goods have better fitting effect, hits on February 18 are predicted to be 570, is expected to be. The establishment of the model is useful for merchants to analyze the online traffic. Upon the forecast data, The inventory of goods can be controlled effectively, the product pages and the defects of goods promotion will be optimized, consequently product conversion rate will rise.

Acknowledgements

Research on Cultivating Students' Professional Skills of E-commerce in Higher Vocational Institutes (TP170109)

References

[1] Li Zhang, Peng Luo. Application of Exponential Smoothing Model to Predict Pearl Production in Japan. Journal of Marine Science. 33(2009): 61.

[2] Weizhong Zhang, Guangzhi Yin, Jianxin Tang, Qinrong Kang. Exponential Smoothing Technique Application in Chongqing Coal Demand Forecast. Journal of Chongqing University (Natural Science Edition).29 (2006): 110.

[3] Zhenping Ma, Weifang Ma. Learn to Use Opportunely Excel 2007 Statistical Analysis Example. Electronic Industry Publishing House, China, 2007.

[4] Chuanhua Yu. Excel Statistical Analysis and Computer Experiments. The Electronic Industrial Press, China, 2009.

[5] Wangeng Chen, Mei Yuan. Application of a Preliminary Study on Exponential Smoothing in Guizhou Province Coal Production Forecast. Journal of Coal Technology. 31 (2012):231.

[6] Xiaohua Wu. Excel in the Application of Exponential Smoothing Parameter Optimization. Journal of Anhui University of Technology (Social Science Edition), 24 (2007): 38.