Advance Journal of Food Science and Technology 7(0: 80-84 205 DOI:0.9026/afst.7.988 ISSN: 2042-4868; e-issn: 2042-4876 205 Maxwell Scientific Publication Cor. Submitted: October 7 204 Acceted: December 8 204 Published: Aril 05 205 Research Article Research on Evaluation Indicator System and Methods of Food Networ Mareting Performance Xinwu Li and Jiaia Zhang Deartment of Electronic Business Jiangxi University of Finance and Economics Nanchang 33003 China Abstract: The research on networ mareting erformance evaluation including evaluation indicator system and methods lies in the core status in mareting erformance management system. A new evaluation indicator system and wavelet BP neural networ algorithm are resented to evaluate food networ mareting erformance. First the evaluation indicator system of food networ mareting erformance is constructed based on analyzing the unique characteristics of food networ mareting erformance; Second the wavelet BP algorithm is designed then genetic algorithm is used and the calculation flows of the new algorithm are redesigned to imrove the convergence seed of wavelet BP neural networ algorithm; Finally the resented evaluation indicator system and algorithm are realized to evaluate networ mareting erformance of three food networ enterrises and the evaluation results show that that the algorithm can imrove calculation efficiency and evaluation accuracy when used ractically and can be used for evaluating other comlicated systems also. Keywords: BP neural networ algorithm food networ mareting erformance evaluation genetic algorithm mareting erformance management wavelet INTRODUCTION As a new attern of mareting networ mareting is becoming a mainstream mareting in all wals of life. Both the raid develoment of the food industry and networ mareting ointly romotes the extensive alication of food networ mareting and networ mareting has gradually become the focus of the food mareting in new century. But as an indisensable comonent of mareting management erformance evaluation needs future consideration and imrovement in evaluation indicator system and methods design in alication of food networ mareting so the research on evaluation indicator system and methods for food networ mareting has become a hot toic for the researchers related and food enterrises (Toher and Sette 20. U to now mathematical models adoted by evaluating networ mareting erformance mainly include the follong categories: Analytic hierarchy rocess is a good method for quantitative evaluation via quantitative method having the functions of establishing the ideal weight structure of evaluated obect value and analyzing the weight structure of actually-built value by evaluated obect; however the method has strong limitations and subectivity th large ersonal error not suitable for comlicated system th lots of evaluation indicators (Yueh 203. Fuzzy comrehensive evaluation is a method carrying out comrehensive evaluation and decision on system through fuzzy set theory the greatest advantage of which is that it wors well on system evaluation of multi-factor and multi-level comlicated roblems. However the membershi of fuzzy evaluation method as well as the definition and calculation of membershi function are too absolute difficult to reflect the dynamics and intermediate transitivity of evaluation indicators of English course education erformance (Ya and Janst 202; Gandha and Bartlett 20. BP neural networ evaluation method maes use of its strong caability in rocessing nonlinear roblems to carry out evaluation of English course education erformance; the method has advantages lie self-learning strong fault tolerance and adatability; however the algorithm is easy to be traed into defects lie local minimum overlearning strong oeration secialization (Thomson 202; Rian and Merve 203. Taing advantage of the ositive effects of BP algorithm the aer overcomes the negative effects of original BP algorithm based on the wavelet algorithm to imrove the woring rincile of ordinary BP neural Corresonding Author: Xinwu Li Deartment of Electronic Business Jiangxi University of Finance and Economics No. 69 East Shuanggang Road Nanchang 33003 China This wor is licensed under a Creative Commons Attribution 4.0 International License (URL: htt://creativecommons.org/licenses/by/4.0/. 80
Adv. J. Food Sci. Technol. 7(0: 80-84 205 networ algorithm to resent a new wavelet BPNN algorithm for evaluating food networ mareting erformance. MATERIALS AND METHODS Evaluation indicator system construction: Evaluation indicator system construction is the first ey ste for evaluating networ mareting erformance correctly and effectively so does the food networ mareting erformance evaluation. And the indicator system for evaluating networ mareting erformance is constituted of multile elements and is a comlex dynamic timely comrehensive system in which various subsystems and indicators coexist in different forms. Base on analyzing the unique characteristics of food networ mareting erformance combined th literatures and exerts consultations (Toher and Sette 20; Yueh 203 and according to connotation characteristics of common networ mareting erformance evaluation. This study constructs a scientific and de evaluation indicator system for evaluating food networ mareting erformance and the system includes 3 hierarchies 3 categories 3 firstgrade indicator 9 second-grade indicator 24 thirdgrade indicator (Table. Designing wavelet BP neural networ algorithm: The algorithm structure and exression of wavelet neural networ and BP neural networ are always the same which consist of inut layer hidden layer and out layer. The ey difference of the two algorithm is that wavelet neural networ taes wavelet transformation function as its excitation function and that of BP neural networ is Sigmoid function. The woring rincile of wavelet neural networ model is that it changes waveform and the scale of wavelet bases continuously to adust the weights and threshold of the networ taing advantage of the rincile of minimum error function. The toological structure of wavelet neural networ model can be seen in Fig. (Sue and Raman 200; Shih 203. f (i means the i th inut and O means the th outut of wavelet neural and networ model resectively and y ( means the th outut of wavelet layer in the above equation i (= 2... inut (= 2... hidden (= 2... outut w i is the connection weights between the i th inut and the th wavelet element. And v is the connection weights between the th wavelet element and th wavelet layer M is the th corresonding sliding scale of wavelet neural and networ model L is the th corresonding contraction coefficient F means wavelet function Φ means the function of outut layer. Follong are the relation between inut and outut of different layers. The relation between inut and outut in the wavelet layer can be seen as Eq. ( and (2 (Sue and Raman 200: inut net = vi f ( i = 2... Inut ( i= y( = F( f ( net = 2... Inut (2 The relation between inut and outut in the outut layer can be seen as Eq. (3 and (4: Hiddenoutut net = w y = 2... Hidden i= (3 Table : Evaluation indicator system for food networ mareting erformance Target hierarchy First-class indicator Second-class indicator Third-class indicator Performance of food networ mareting Website erformance side Website design Domain name choice Style and visual effects Retrieval function Information udate frequency Website roerties Safety Functional comrehensive Download seed Lin availability Interactive convenience th users Website oularization Effects of search engine romotion Effect of networ advertisement Website flow The number of visitors The average number of age views Flow conversion rate Enterrise erformance side Financial erformance Sales revenue growth rate Sales rofit growth rate Cometition erformance Brand awareness Maret share growth Consumer ermeability Customer erformance side Customer service Service resonse seed Problem solving efficiency Customer satisfaction Food logistics distribution 8 Customer feedbac evaluation Received the integrity of goods The seed of food receit
Adv. J. Food Sci. Technol. 7(0: 80-84 205 fit ( i = / E( i (6 E( i 2 K = = = ( d a2 (7 In which d means target outut and a2 means the actual outut of networ. Fig. : The toological structure of the imroved algorithm y( = Φ( f ( net = 2... Hidden (4 From Eq. (3 and (4 it can get the relation model between inut and outut of wavelet neural networ easily lie Eq. (5: Evolution comutation: Evolution comutation is conducted through selecting oerator and is a rocess of oulation generational renewal. Using fitness of each chromosome the algorithm selects the chromosome of next generation from oulation based on samling mechanism of fixed selection robability. The evaluation tendency of the oulation is decided by the characteristics of selection oerator which is always be described through three asects of samling sace samling mechanism and selection robability. Hidden Inut = Φ( v F(( = i= O w f ( i M / L i Imroving wavelet neural networ algorithm th genetic algorithm: Chromosomes exression and initial oulation generation: The training rocess of neural networ is the rocess of determining the weights there sholds and telescoic translation oerators so it should taes weights IW LW and threshold b b 2 and telescoic translation oerators a b as decision variables in training neural networ th genetic algorithm and then maes decision variables as a letter string and taes it as a solution to the roblem. The algorithm uses real number encoding the string is as follows. IW IW 2... IW i... IW JI LW LW 2...... LW KJ a a2... b b... b 2 J b J b2 b2... b2. In which 2 IW i means the connection weights between the i th neuron in inut layer and the th neuron in hidden layer LW means the connection weights between the th neuron in hidden layer and the th neuron in outut layer a J means the exansion and contraction arameters of the th neuron b J means the translation arameters of the th neuron b J means the threshold of the th neuron and b2 means the threshold of the th neuron in outut layer. Initial oulation rocess includes determining oulation size cross robability and secifying a random value interval of IW and LW of each genes and then generates a new gene. (5 Target function and fitness function: Genetic algorithm only can be evolved along th the increase direction of fitness function increase value so the fitness function can be designs as recirocal form of obect function see as Eq. (6 and (7: 82 Genetic comutation: Genetic comutation includes cross oeration and mutation oeration. And cross oeration conducted here occurred in the same loci of different two chromosomes not means the concrete chromosomes. Cross oeration rate means the ration of genes involve in crossover oeration to total numbers of genes. Weights value breas the value sace of initial weights though mutation oeration and searches toward a more extensive sace. Imroving adative method of heromone evaoration intensity ρ: Pheromone evaoration intensity ρ can affect the global search erformance of genetic algorithm directly. If ρ value is too small the cumulative heromone intensity in the selected road ll be too large and easy to early-maturing and If ρ value is too large the cumulative heromone intensity in the unselected road ll be evaorated and ll be become small and that maes global search erformance of genetic algorithm decreased. If the ρ has no imrovement after many circulation calculation times Eq. (8 can be taen as adative method of Pheromone evaoration intensity ρ: ex( α ρ( t ρ( t + = ρmin if ex( α ρ( t ρmin (8 else According to exeriment test it ll has referable calculation effects when the value of α lies in the sace of (0.000 0.00. The value of ρ min cannot be too small also and tae it equal to 0.2 in the study. In order to highlight the mechanism that how the otimal solution of last generation attract the next generation the next generation of attractive highlight on the generation of otimal solution Eq. (9 can be used to udate heromone: num τ ( G = ( ρ ( τ G + ( τ G (9
Adv. J. Food Sci. Technol. 7(0: 80-84 205 Table 2: Part evaluation results of different networ enterrises Website erformance Enterrise erformance Customer relationshi Final evaluation Cororation A 3.07 3.298 3.64 3.234 Cororation B 3.47 3.90 4.279 3.803 Cororation C 4.002 4.508 4.793 4.408 Table 3: Evaluation erformance comarison of different algorithms Algorithm Algorithm in the study Ordinary BP algorithm Fuzzy evaluation algorithm Accuracy rate 93.98% 83.98% 69.28% Time consuming (sec 56 In which only num ants find their otimal solution and the global heromone can be udated by these num num ants; nnnnnn ττ (GG wwww means the heromone summary of the ants that got their otimal solution in this circulation calculation (Wettler 202; Tuffer and Wooder 20. Imroving the convergence seed of the algorithm: BFGS (Broyden Fletcher Goldfarb and Shanno algorithm is a very effective otimization algorithm esecially its sueriorities in solving high dimensional otimization roblems comared th gradient descent methods. So the aer uses BFGS algorithm to imrove the seed u convergence seed of original wavelet neural networ algorithm and remedies the disadvantages of too long search time of original genetic algorithm. The aer taes relative average outut error equation of wavelet neural networ as T obective function suose W ( w w2... as weight vector of wavelet neural networ its udating equations can be seen as Eq. (0 to (2: H + = H + H G W H / G T T µ (0 T T investigation. In order to mae the selected data reresentatives 300 consumers (00 consumers from each cororations th more than one and half years networ food buying exerience 300 consumers th one year food buying exerience 300 consumers (00 consumers from each cororations th less than year food buying exerience. In order to save aer sace here omits the intermediate evaluation results only final comrehensive evaluation results and some secondary evaluation results and rovided (Table 2. As for the evaluation erformance of the advanced algorithm the ordinary BP neural networ (Rian and Merve 203 and fuzzy evaluation (Gandha and Bartlett 20 are also realized in the aer and the evaluation erformance of different algorithms can been seen in Table 3. In Table 3 evaluation results of different 3 food networ enterrises are chosen and comared th artificial evaluation to calculate the evaluation accuracy. The indicators of the calculation latform can be listed as follows Intel i3 220 2 GB DDR3 AMD Radeon HD 7450 and 3.3 GHz CPU C rogramming language environment and ndows XP oerating system. W = W β H E ( W ( + T T µ = + ( G H G /(( H G (2 In which W = W + -W G = E( W - E( W means the number of training suose P = H E( W β can be gotten by adative method and should satisfy the Eq. (3: E ( W + βp = min E β 0 ( W + βp RESULTS AND DISCUSSION + (3 Taing exerimental data 3 food networ cororations to create exerimental database the three food networ enterrises are called cororation A B and C resectively. For customer art data 300 consumers of each food networ cororations are investigated and the results is taen as the database for training and exerimental verification in the study totally 900 consumers data for exerimental confirmation which come from ractical visit and 83 CONCLUSION At resent evaluation of networ mareting erformance esecially for food networ mareting erformance evaluation lies in the core status in mareting erformance management system and is an effective way to guarantee networ mareting erformance for networ trade enterrises. So the research on networ mareting erformance evaluation including evaluation indicator system and methods has become a hot toic for the researcher related. Taing advantage of the ositive effects of BPNN and wavelet algorithm the aer overcomes the negative sides and constructs a new evaluation indicator system for evaluating food networ mareting erformance. The case study in the aer taes the data of three food networ cororation as an examle to illustrate the referable evaluation effects of the resented evaluation indicator system and method. ACKNOWLEDGMENT This study is suorted by the scientific research roect of the education deartment of Jiangxi Province
Adv. J. Food Sci. Technol. 7(0: 80-84 205 (GJJ3300 and the 52 nd Chinese Postdoctoral Fund under the grant (202M52284. REFERENCES Gandha S. and L. Bartlett 20. Categorizing web features and functions to evaluate commercial web sites. Ind. Manage. Data Syst. 7(: 854-862. Rian V. and J. Merve 203. A framewor and methodology for evaluating E-commerce web sites. Electr. Netw. Al. Policy 6(2: 23-237. Shih S. 203. Quantitative assessment of Euroean municial web sites: develoment and use of an evaluation tool. Internet Res. 7(: 28-39. Sue T. and D. Raman 200. A new BP algorithm for evaluating internet mareting system erformance. J. Qual. Manage. (4: 925-933. Thomson S. 202. Usage and effectiveness of online mareting tools among business-to-consumer (B2C firms in Singaore. Int. J. Inform. Manage. (4: 42-430. Toher S. and Z. Sette 20. Analyzing the networs mareting erformance based on fuzzy information. J. Bus. Manage. 0(0: 30-30. Tuffer S. and S. Wooder 20. An imroved wavelet algorithm based on factor analysis. J. Inform. Manage. (3: 09-8. Wettler S. 202. Research on wavelet-based system evaluation model. Int. J. Manage. Sci. 2(5: 7-28. Ya Y. and S. Janst 202. Fuzzy quality attributes for evaluating internet mareting system erformance. Total Qual. Manage. 0(2: 93-936. Yueh Y. 203. Otimal model of comlicated system evaluation based on linear weighting. Ind. Eng. J. 8(9: 77-87. 84