Evaluation of e-Commerce Websites using an Optimized RBF

Algorithm based on Fruit Fly Algorithm

Ren Changrong

Electronic Commerce Department, Liaoning University of International Business and Economics,

Dalian, Liaoning, 116052, China

Keywords: Fruit Fly Algorithm, RBF Algorithm, e-Commerce Website, Optimized Algorithm, Evaluation.

Abstract: Developments of e-commerce modes have brought in increasingly fierce competition. The RBF algorithm

can perfectly and accurately evaluate e-commerce websites, thus providing quantified competitive index of

e-commerce. Furthermore, evaluation can be more accurate by using an optimized RBF algorithm based on

fruit fly algorithm. With this optimized RBF algorithm, the work accurately calculated the evaluation results

of e-commerce websites, thereby resolving the competition issues of e-commerce sites fundamentally.

1 INTRODUCTION

Developments of the Internet have led to a rise of e-

commerce websites and brought in fierce

competitions among these websites. In this context,

e-commerce websites hold a very important position

in economy. The major topic of current discussion is

how to quantify the evaluation of e-commerce

websites. The RBF algorithm is the general method

to do the evaluation. However, due to some

impracticabilities and defects, this method needs to

be optimized. Therefore, an optimized RBF

algorithm based on fruit fly algorithm is proposed.

This new algorithm has been applied into the

evaluation of e-commerce websites.

2 DEFINITION OF THE RBF

ALGORITHM

RBF, known as the radial basis function, is a single

contact feedforward neural network. RBF has been

widely used in pattern recognition and signal

processing. Results have shown this method is

feasible. Advantages of RBF are obvious: simple

network structure and strong nonlinear

approximation ability. In practice, center parameters

of neurons need to be identified when using RBF

algorithm. Center parameters can be selected in two

ways: directly select from machine training samples,

or use clustering method. Both methods have flaws,

thus the application of RBF algorithm is limited.

3 STRUCTURE OF THE RBF

NEURAL NETWORK

ALGORITHM

RBF neural network consists of three layers: input

layer, middle layer (also known as hidden layer) and

output layer. Detailed analyses for each layer are

made to study the structure of RBF neural network

algorithm further.

3.1 Input Layer of the RBF Neural

Network

The input layer, including the initial source points, is

the entrance of data. Composed of sensing units,

source points are connected with external neural

network. Their primary function is message passing.

Messages are directly transferred into input layer

without any transform. Such input layer used to

collect data can be found in any neural network.

3.2 Hidden Layer of the RBF Neural

Network

When messages are collected by input layer, they

will be sent into hidden layer. Hidden layer is the

processing layer of the RBF neural network. In this

layer, input message will be subjected to non-linear

transformations by RBF, and therefore they can have

higher dimensions. In mathematical perspective,

hidden layer is the processing unit of the RBF neural

312

Changrong R.

Evaluation of e-Commerce Websites using an Optimized RBF Algorithm based on Fruit Fly Algorithm.

DOI: 10.5220/0006025003120315

In Proceedings of the Information Science and Management Engineering III (ISME 2015), pages 312-315

ISBN: 978-989-758-163-2

network, where combines all core parts and realizes

the internal conversion of input data.

3.3 Output Layer of the RBF Neural

Network

When data are transferred from input layer to hidden

layer, core module functions will perform non-linear

transformations on data to make them linearized.

After a series of data processing, linear data are

generated. These data are sent to output layer and

can provide input layer with activation responses.

4 ALGORITHM DESIGN OF THE

RBF NEURAL NETWORK

ALGORITHM

Algorithm design is a network-based design

approach. In some applications, the RBF algorithm

design should achieve quantification and

linearization. Generally, algorithm design has two

steps: the first is to determine the kernel function of

hidden layer, which is the RBF; the second is to

adjust the RBF. By analysis, RBF will not affect the

performance of network. RBF is called in the form

of nonlinear function, and then can be used to

determine the central value and width, adjust the

algorithm and so on.

4.1 Network Algorithm Design based

on the RBF Neural Network

Algorithm

During the process of network algorithm design, the

neural network design can be divided into following

steps:

The first is to construct the neural network. The

neural network works as the general framework.

Key points in its construction are to determine the

different layers and its overall structure.

Specifically, the source of sample data from input

layer can determine the network structure.

Moreover, middle layer and output layer can

determine neurons and kernel functions.

The second is the initialization process. When

network is constructed, different data are input to it

according to demands. After weight values are

initialized, initial data can be obtained.

The third is the internal training. After

initialization, the network needs to be trained by

input and output tests with sample data.

The fourth is simulation. After the completion of

the above, network simulations can be processed.

4.2 Design Flow of Neural Network

Algorithm

When constructed the infrastructure of neural

network, the design flow of this algorithm should be

considered. This work contributes to obtain

corresponding network architecture and perfect test

environments. The first step is to select the cluster

center, whose value determines the initial width. The

input data serve as distance criterion of center layer,

and their volume determines the number of

clustering in the network. The second step is to

update weight values. Updated values are

determined by the instantaneous values of certain

functions with the input of all samples.

5 OPTIMIZED RBF NEURAL

NETWORK BASED ON FRUIT

FLY ALGORITHM

Many studies have been done on the structures and

composition simulations of RBF neural network. In

practice, one problem is that post optimizations are

needed due to some limitations of RBF neural

network. However, the fruit fly optimization

algorithm (FOA) has improved the RBF neural

network algorithm.

5.1 Fruit Fly Optimization Algorithm

FOA is a calculation method proposed by a teacher

in Taiwan. This method calculates in an evolutional

way to achieve certain purposes. When fruit flies

look for their food, they use their keen sense of

smell and vision to quickly locate the food. FOA is a

global optimization method derived from the way

fruit flies forage.

5.2 RBF Neural Network based on

FOA

FOA is a calculation method derived from the

process that fruit flies forage. In practice, FOA

requires a step decomposition to achieve its goals, so

it can be divided into following steps:

(1) Group locations of fruit flies should be

initialized and then recorded.

Evaluation of e-Commerce Websites using an Optimized RBF Algorithm based on Fruit Fly Algorithm

313

Evaluation of e-Commerce Websites using an Optimized RBF Algorithm based on Fruit Fly Algorithm

313

(2) The distance of random search for fruit flies

needs to be determined. Random location data can

be calculated by certain functions.

(3) Random directions of fruit flies’ motion

represent the possible directions of food. In this

experiment, because the position of food is

unknown, the reciprocal of distance will be

estimated by the distance between fruit flies and the

base point.

(4) Fruit flies search their food according to odor

concentration. Therefore, the decision function can

be obtained by calculating the odor concentration.

(5) The optimal concentration of food odor is

determined by the decision function.

(6) The food location can be determined

according to the calculated group location data of

fruit flies.

6 E-COMMERCE WEBSITE

EVALUATION USING AN

OPTIMIZED RBF ALGORITHM

BASED ON FOA

Optimized RBF neural network based on fruit fly

algorithm has overcome the shortcomings of RBF

neural network. Using optimal locations of FOA is

the best way to construct a RBF neural network.

Now the question is: how to use optimized RBF

neural network based on fruit fly algorithm to

evaluate e-commerce websites?

6.1 Competitive Index Evaluations of

e-Commerce Websites using an

Optimized RBF Neural Network

based on Fruit Fly Algorithm

The analytic hierarchy process(AHP), a common

used analytical method, applies to evaluations of e-

commerce websites. AHP is proposed by an

American scientist. It is a hierarchical and structured

analysis method combined with quantitative analysis.

AHP is a universal method that can effectively solve

practical problems. Therefore, AHP can be used to

evaluate the e-commerce website.

6.2 Data Determination in Input and

Output Layers of RBF Neural

Network

Input layer is a crucial port in RBF neural network

to connect to external data. To determine the input

data of e-commerce websites, the competitive index

needs to be calculated. Input data which need to be

determined can be scored by calculating the

expectation. Scores can largely reflect the

competitiveness of these e-commerce websites. In

this way, the reference value of input data will be

more reliable. This process is equivalent to the data

initialization in FOA. In the later calculation process,

the best data can be obtained by referring the input

data and using the designed network.

6.3 Competitive Index Calculation

using AHP

AHP obains its corresponding data by calculating

different evaluation indexs. Weight values are key

indexs in the calculation process, in many cases, the

level of weight values will directly affect the final

results. To shorten the testing process, kernel

functions will perform nonlinear conversions in

different layers .

6.4 Application of Competitive Index

Evaluations of e-Commerce

Websites using RBF Neural

Network Algorithm

When use the RBF neural network to evaluate the

competitive index of e-commerce websites, the

process can be divided into the following steps.

The first is to input the competitive index data.

Experts’ evaluations can be used as reliable initial

data.

The second is to use these data to create a sample

database, this database will work as the initial

reference. Neural network is also established by

these data, then researchers will conduct tests and

simulation trainings on it.

The last step is do the calculation with data

models, so as to obtain the competitive index

standard of e-commerce websites.

7 CONCLUSIONS

This work analyzed the functions of different layers

in RBF neural network design. The output of RBF

neural network should not be affected by input layer

and nonlinear transformations. To meet this

requirement, researchers can set corresponding

parameters in different input ports. In practice, FOA

are used to overcome the limitations of RBF neural

network. So RBF neural network based on FOA are

ISME 2015 - Information Science and Management Engineering III

314

ISME 2015 - International Conference on Information System and Management Engineering

314

adopted in the evaluations of e-commerce websites.

Specifically, nonlinear data extracted from websites

are conversed in middle layer, and eventually they

transformed into linear and quantified data. FOA can

be used to find the optimum point, and therefore the

most authentic quantitative data of e-commerce

websites are obtained. In conclusion, it is effective

for people to use RBF neural network based on FOA

to do evaluations of e-commerce websites.

REFERENCES

Dou Yutan, Wang Daxing, Liu Huaqing , et al. Methods of

Constructing Virtual Well Using FOA in Well-lacked

Area of Huangtuyuan. Progress in Geophysics,

2014(02):674-678.

Huang Jiangtao, Gao Zhenghong, Bai Junqiang, et al.

Research on Applications of RBF Radial Basis

Function and the Delaunay Graph Mapping

Technology in Vehicle Frame Shape Design. Acta

Aerodynamica Sinica, 2014(03):328-333.

Wang Cuiping, Zhen Ya, Zhen Jinjing, et al. Wolbachia

Alike Male Defects Induced Cytoplasmic

Incompatibility Caused by Knock Down of Fruit Fly

Mst84Db. Acta Entomologica Sinica, 2014(1):92-94.

Evaluation of e-Commerce Websites using an Optimized RBF Algorithm based on Fruit Fly Algorithm

315

Evaluation of e-Commerce Websites using an Optimized RBF Algorithm based on Fruit Fly Algorithm

315