Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
The embodiment is a method for planning electronic cigarette product lines by considering market factors, which comprises the following steps:
step 1: determining the attributes of the electronic cigarette product considering market factors;
the market factors include the type of the electronic cigarette and the product attributes of the electronic cigarette.
Step 1-1: determining the type of the electronic cigarette;
the Taobao, the Jingdong and various electronic cigarette forums are investigated, and the electronic cigarettes sold in the market can be divided into four types: smoking cessation type electronic cigarettes, cigarette replacement type electronic cigarettes, cigarette-like type electronic cigarettes and high-grade gift type electronic cigarettes.
Step 1-2: collecting commodity comments by utilizing a web crawler technology;
and crawling the commodity comments of the Jingdong website as an analysis sample by using a crawler technology to finally obtain about 24 thousands of comments.
Step 1-3: analyzing the commodity comment by utilizing a word segmentation technology;
step 1-3-1: segmenting words to obtain candidate characteristic words;
the comments are subjected to word segmentation and cleaning, and nouns with frequency greater than 20 times are obtained as shown in tables 1 and 2, wherein only two kinds of feature words of the electronic cigarette are displayed, table 1 represents candidate feature words of the smoking cessation type electronic cigarette, and table 2 represents candidate feature words of the cigarette replacement type electronic cigarette.
TABLE 1 candidate characteristic word (1)
TABLE 2 candidate characteristic words (2)
Step 1-3-2: calculating the importance of the customer requirements;
the importance of customer demand is calculated by dividing the number of occurrences of a certain feature word by the total number of occurrences of all feature words.
Step 1-4: converting the product characteristics to product attributes by using a QFD technology;
QFD is a multilevel deductive analysis method for converting the requirements of consumers into design requirements, part characteristics, process requirements and production requirements, and is used for guiding the robust design and quality assurance of products;
summarizing fourteen customer demands which may be concerned by customers according to feedback of user use information, wherein a related relation matrix of the QFD quality house is given in a table 3, and the correlation between product characteristics and product attributes is represented by numbers, wherein 0 represents no relation, 1 represents that the relation is not tight, 3 represents that the relation is general, 5 represents that the relation is tight, and 2 and 4 represent intermediate relations of 1-3 and 3-5 respectively;
TABLE 3 relationship between customer demand and engineering characteristics
By using a matrix analysis method, the weight of the product characteristics is converted into the weight of various product attributes meeting the requirements of customers, so that the product attributes required to be planned for the smoking cessation type electronic cigarette and the cigarette replacement type electronic cigarette are obtained, and after the detailed attribute levels are divided for the product attributes, the specific information shown in tables 4 and 5 is obtained. Wherein, table 4 gives the product attribute and attribute level of the smoking cessation type electronic cigarette, and table 5 gives the product attribute and attribute level of the cigarette replacement type electronic cigarette;
through the steps, the number of the product attributes needing decision making is reduced from 18 to 9, the burden of the respondents is reduced, and meanwhile, the prediction accuracy is improved.
TABLE 4 horizontal relationship of the product attributes of the electronic cigarette (1)
TABLE 5 horizontal relationship of the product attributes of the electronic cigarette (2)
Step 2: establishing an electronic cigarette product line optimization model;
suppose that the H enterprise is to develop new product lines to meet the actual needs of different customer segments. The product line comprises a fixed number of products, the number of attribute levels of each attribute in the products is fixed, and the attribute levels of the product attributes are different, so that the products have similar functions and slightly different performances;
through the previous market research and the customer demand analysis, the customers in the market corresponding to the target product line of the company H are clustered into I market segments, wherein each market segment comprises ni(I1, 2.., I) prefer similar customers, while the market has M competing products. The perceived utility surplus based on the currency amount is used to express the importance level of each product attribute to the customer, namely the customer perceived utility minus the price of the product variant. Thus, in a profit-oriented optimization model, price is a decision variable. The invention takes the price of the product variant and the specific product configuration as main decision variables to achieve the aim of maximizing the total profit of the product line;
The symbols involved in the model are explained below:
i number of market segments;
number of product variants;
nirepresenting the number of customers in segment i;
Pija likelihood that a jth product variant is selected in an ith market segment;
k number of product attributes;
Lka horizontal number of kth product attributes;
Ckl(ii) a variable cost corresponding to the l level of the kth product attribute;
Cffixed cost of product development;
Cvvariable cost of product development;
the total cost of product development;
scaling factor of μ MNL rule;
Uijutility value of jth product in ith market segment;
uiklthe kth product attribute of the product, the l attribute level is the component utility value of the ith market segment;
γijrepresenting the utility remaining value of the product variant j in the market segment i;
representing the utility surplus value of the competitive product j in the market segment i;
m number of competing products;
g total market profit of the product line;
r total market revenue for product line;
wjka weight of a kth product attribute for a jth product;
the symbolic representation of the decision variables is:
xjklis a variable from 0 to 1 if the l < th > attribute level of the k < th > attribute of the product variant j is selectedThe value is 1, otherwise, the value is 0;
pjprice of jth product variant.
Step 2-1: analyzing the preference of the customer by using a joint analysis method;
the most common method for analyzing consumer preferences is a joint analysis method, which explores the influence of different product attribute configurations of products on purchasing decisions of customers and obtains a joint Preference Structure (Additive joint Preference Structure) by using a statistical analysis technique. When analyzing the preference of a customer by means of a joint analysis method, firstly, the customer needs to evaluate a group of virtual products with different attribute level combinations in a scoring mode, and then evaluation results obtained by the investigation are processed by using methods such as least square regression and hierarchical Bayes, so as to estimate the component utility value of each product attribute level;
in the joint analysis method, there are many methods for calculating the utility value of the attribute-level component, and according to the most common component utility analysis model, the overall utility value of a product variant can be regarded as the linear superposition of the utility values of each component, and the formula is as follows:
generally, the purchasing behavior of a customer for a product variant depends mainly on the remaining utility perceived by the customer for the product variant. The utility residual (in currency) for a product variant is the difference between the utility value (in currency) of the product and its sales price, and the formula is as follows:
γij=Uij-pj,i=1,2,...,I;j=1,2,...,J
step 2-2: analyzing the purchasing behavior of the customer by utilizing MNL customer selection rules;
the purchase behavior of the consumer for a new product is influenced by the surplus of the utility of the product itself as well as the surplus of the utility of the existing like product. In recent years, probabilistic selection rules have been applied many times because such rules are more realistic and reliable in describing consumer purchases. Among the probabilistic customer selection rules, the most common is the polynomial logarithm MNL selection rule, whose formula is as follows:
where μ is a scaling factor, if the value of μ is very large, the model is similar to the deterministic rule; if the value of μ is approximately zero, the model is similar to a uniform distribution. The value of μ can be corrected by the results of the actual market share investigation.
Step 2-3: analyzing the cost of the product line;
the cost of the product line of the present invention is divided into two parts, variable cost and fixed cost, so the total cost of the enterprise product line is as follows:
C=Cf+Cv
wherein, CvRepresenting variable costs of the product line, CfRepresenting the fixed cost of the product line. The fixed cost of the product line comprises the cost of a company for deciding to construct a product line and the project starting cost, the management cost, the capital construction cost and the like for deciding to produce the product; in addition, the variable cost of a product line includes the costs associated with the production of the product by the enterprise, such as packaging costs, assembly costs, storage costs, and logistics costs. According to a Linear-additive CostModel (Linear-additive CostModel) and a product utility measurement method, variable cost is expressed as a contribution degree of an attribute level to cost, so that a cost function is as follows:
step 2-4: establishing an optimization model;
step 2-4-1: an objective function;
the optimization goal of the present invention is product line profit maximization. The fixed cost of the product line has no relation with any decision variable, and the result of the objective function is not influenced, so that the result is ignored, and the following formula is the final objective function:
step 2-4-2: a correlation constraint;
the necessary constraints involved are as follows;
(1) for each product attribute in a product variant, there is and can only be selected one attribute level, and this constraint can be expressed as the formula:
(2) the configuration of any two product variants in a product line cannot be exactly the same. This relationship can be expressed as the formula:
step 2-4-3: optimizing the model;
finally, establishing a complete nonlinear product line optimization model:
and step 3: designing an intelligent optimization algorithm for solving the model;
the flow chart of the overall algorithm is shown in fig. 2, and the specific steps are as follows.
Step 3-1: designing a chromosome structure and a coding mode;
in the chromosomes designed by the invention, one chromosome represents one product line developed by an enterprise, each chromosome is composed of J subregions, the jth subregion contains information related to the attribute configuration of the jth product variant, each subregion comprises K +1 genes, the kth gene contains information related to the kth attribute of the product and represents the attribute level selected by the attribute, the K +1 gene represents the price of the product variant, because the actual product selling price is usually an integer, the efficiency can be improved by carrying out integer coding on the selling price of the product variant, and the process of dispersing the price interval into several integers can be described as follows: firstly, analyzing the investigation result of a consumer to obtain the component utility value of each attribute level of the product variant; then, estimating the price range of the product variant through the component utility value and the cost of each attribute level of the product variant; and finally, discretizing the obtained price interval to obtain W integers. Thus, the gene representing the selling price of the product variant can be selected from among W integers;
an example of chromosome coding is given in figure 3. There are two product variants in this product line, each with three attributes from the consumer's demand perspective. According to the encoding rule, the attribute configuration of the product 1 (sub-partition 1) is as follows: the 1 st gene represents that the attribute 1 of the user demand angle selects the 1 st attribute level, the 2 nd gene represents that the attribute 2 of the user demand angle selects the 2 nd attribute level, the 3 rd gene represents that the attribute 3 of the user demand angle selects the 3 rd attribute level, and the 4 th gene represents that the price of the product 1 is the 6 th discretization price. The encoding process for product 2 is similar to that for product 1 and need not be repeated.
Step 3-2: initializing a population;
to maintain population diversity, chromosomes in the initial population are randomly generated. The attribute configuration of the product is expressed by integer coding in the gene, and the generation ranges of the random functions of the attribute level of the user demand angle and the index value of the product price are [1, L ] respectivelyk]And [1, W]。
Step 3-3: designing a genetic operator;
step 3-3-1: designing a crossover operator;
the present invention employs a Uniform Crossover process (UCM). Fig. 4 shows an example of a crossover operation performed on a chromosome. Firstly, randomly generating a binary cross mask for selecting whether the corresponding position of a father chromosome is subjected to cross operation, and if the value of a certain bit of the mask is 0, indicating that the corresponding gene bits of the father chromosome and the mother chromosome are exchanged; if the value corresponding to the cross mask is 1, no swap is performed. The mask in this example is 10101110, and according to the intersection rule described above, for the parent and the mother, the attribute level and the price index value of the 2 nd attribute of product 1 and the price index value of product 2 need to be exchanged respectively, and the values of the other loci of the two chromosomes do not need to be exchanged, resulting in child individual 1 and child individual 2 as shown in fig. 4.
Step 3-3-2: designing a mutation operator;
the invention uses a random variation mode, namely if a certain chromosome needs variation under a given variation rate, some gene positions of the chromosome are selected, and the value of the gene positions generates a new legal gene again at random within the value range.
Step 3-3-3: a chromosome repair strategy;
there may be identical parts in the chromosome structure after the crossover mutation operation, i.e. there are products in the product line with identical product configurations. In the actual product line optimization design, products of the same product line should be configured differently, so in this case, chromosome repair is required, and infeasible solution is feasible. The following describes the steps of the chromosome repair strategy used in the present invention;
step 3-3-3-1: judging whether the gene values (attribute levels) corresponding to the corresponding units (attributes) of each sub-partition (product) of the chromosome (namely the actual product line) in the current population are completely the same, namely, checking whether the product variant with repeated attribute configuration exists on the same product line;
step 3-3-3-2: based on the existing individual with repeated product attribute configuration, aiming at the repeated product attribute configuration, randomly mutating the value of a certain gene position within the attribute level range of the repeated product attribute configuration to be used as a repaired legal chromosome;
step 3-3-3-3: checking whether repeated attribute configuration conditions still exist in the chromosome at the moment, and if so, switching to the step 3-3-3-2;
step 3-3-3-4: the above operations are repeated until there is no repeated attribute arrangement, and the operation is stopped.
Step 3-4: establishing a fitness function and selecting;
step 3-4-1: a fitness function;
in the present invention, the objective function is taken as a fitness function.
Step 3-4-2: selecting;
the Roulette Wheel Selection (Roulette Wheel Selection) is the method of choice in the present invention. The basic idea of the method is as follows: the probability that a chromosome is selected to be inherited to the next generation is proportional to its fitness, the greater the likelihood of being selected to remain in the next generation. Assume that the size of the population is Popsize, where the fitness of the individual is fpThen the probability P that the individual P is selectedpCan be expressed as the formula:
after the probability of each chromosome being selected is obtained, in order to select mating individuals, multiple rounds of selection are required, and a random number uniformly distributed between [0,1] is generated in each round as a selection pointer to obtain selected individuals. The selection process continues until population size is reached.
Step 3-5: setting a stopping criterion;
the invention defines the stopping criterion by setting the maximum iteration times, namely when the iteration times reach a preset threshold value, the searching process is stopped, and the optimal solution in the current population is output as the final result.
Step 3-6: constructing a genetic algorithm embedded in a hill climbing algorithm;
the local search capability of the traditional genetic algorithm is weak, and meanwhile due to the complexity of an electronic cigarette product line optimization model, if the local search algorithm is considered to be added, the time consumption is strictly controlled, so that the introduced local algorithm is required to be simple and efficient. Based on this, hill climbing algorithms are adopted to improve the algorithm's local search capabilities. The basic idea of this algorithm is to compare the search process to a hill-climbing process, going up a hill in the direction of increasing height without any other information about the crest of the hill. The main flow of the algorithm is as follows:
step 3-6-1: determining an original solution marked as g;
step 3-6-2: initializing a position pos (0) where a local search starts on a chromosome, and indicating a subscript index of a gene segment where the local search starts;
step 3-6-3: if the stopping criterion is not met, namely the pos value is smaller than the length of the chromosome, the step 3-6-4 is carried out, otherwise, the hill climbing algorithm is ended;
step 3-6-4: performing local search at the current pos position, namely making next equal to NEIGHBOR (g, pos), wherein the next represents the neighborhood solution at the moment;
step 3-6-5: if the solution performance of the next is better than that of the g, making g be the next and pos be pos +1, returning to the step 3-6-3, and continuing to perform local neighborhood search; otherwise, let pos ═ pos +1, return to step 3-6-3.
The following will be verified with one case experiment.
The case of the electronic cigarette of the H enterprise is adopted for carrying out experimental analysis, the enterprise plans to develop two electronic cigarette product lines to meet the requirements of customers in different market segments on the electronic cigarette, products produced by each product line are positioned in two market segments of light smokers and heavy smokers, and the number of the customers in the two market segments is respectively 170000 and 40000 estimated by the number of the customer reviews of the E-commerce website. The first product line is used for producing the smoking cessation type electronic cigarette, and the second product line is used for producing the cigarette replacement type electronic cigarette. The number of all attribute levels in the product is fixed, the attribute levels of product attribute selection are different, so that the electronic cigarette products have similar functions and slightly different performances, and enterprises need to select proper attribute levels to construct proper products to form a product line, namely, the attribute configuration of the products is determined, so that the strategic purpose of companies is achieved.
Because the planning methods of the two product lines are the same, only the steps and the specific implementation method of the cigarette-replacing type electronic cigarette product line are introduced.
In the first step, the attributes and attribute levels of the product are given. The attribute levels of the product attributes of the cigarette replacement type electronic cigarette are given in table 5, which are 3, 2, 4, 2, 4, 3 levels in order.
And secondly, combining all attributes and attribute levels of the product by adopting an orthogonal design to generate a virtual product set. Since any combination of individual product attributes and attribute levels is effective, there are a total of 3 × 2 × 4 × 3 — 13824 possible solutions. Because the magnitude order of possible product schemes is too large, 32 representative virtual products are obtained by orthogonal design of IBM SPSS statics software, and the difficulty of market research is greatly reduced. The attribute configuration information for these 32 virtual products is shown in table 6 (the integer in each bin in the table represents the serial number of the selected level).
TABLE 6 orthogonal virtual products for electronic cigarettes
And thirdly, calculating the utility value of the product attribute. And collecting data, obtaining evaluation information of the customer on the virtual product through market research, and excavating preference data of the customer on the virtual product. Before the experiment is developed, a large number of customers are invited to carry out market research, and preference information of the customers on the electronic cigarette products is collected. Wherein, each interviewee needs to evaluate the utility of 32 orthogonally designed virtual products, and the component utility value of each attribute level is obtained by designing a least square linear regression method. Meanwhile, the expert evaluates the cost of the product and gives the fixed cost of the product and the variable cost corresponding to each attribute level. The utility value and cost of each attribute are shown in table 7.
TABLE 7 Utility values and costs for attributes
And fourthly, calculating the utility residual value of the competitive products. Three competitive products exist in the market, and the specific configuration and price of the competitive products are shown in the table 8. On the basis of the third step, the utility remaining value of the competitive company product, namely the difference between the calculated component utility value and the price, can be obtained by combining the product configuration and the price of the existing competitive products in the market, and the calculation result is shown in table 9.
TABLE 8 specific configuration and price of competing products
TABLE 9 Utility residual of competing products
Fifth, a discrete price set of product variants is estimated. To simplify the overall utility value formula for one product variant in step 2-1, assume that the weight variables w in the formula are each set to 1. From the ingredient utility values obtained in table 7, the highest expected utility value for each product in two market segments can be estimated:
from the results, the highest expected utility value of 179.6 for the product variant in segment two is the highest expected utility value in all segments, which is taken as the upper limit for the price of the product variant. The highest part purchase cost for a product variant can be derived from the cost for each attribute in table 7:
the highest purchase cost is C13+C23+C31+C44+C52+C61+C74+C84+C93=162
The calculated highest part purchase cost 162 for the product variant is taken as the lower limit for the price of the product variant. Thus, the selling price of the product variant can be specified as a continuous value between 162 and 179.6. In actual life, the selling price is generally a discrete integer value, so that the continuous selling price interval of the product variant should be discretized, and the obtained discrete price set is as follows: {162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179}.
The experiment randomly obtains an initial solution, each group of experiment can be executed for a plurality of times in order to prevent random factors from influencing the observation of the final change trend of the experiment result, and each initial solution is regenerated. The results were averaged over multiple experiments.
Aiming at the problem characteristics, a genetic algorithm and a tabu search algorithm are used for solving, and an algorithm parameter experiment of the two algorithms and an optimization capability comparison experiment of the two algorithms are carried out:
experiment 1: and analyzing the influence of the size change of the population scale, the change of the cross rate and the change of the variation rate on the average value of the optimal solution aiming at the genetic algorithm.
Experiment 2: and comparing the genetic algorithm embedded with the hill climbing algorithm with the genetic algorithm not embedded with the hill climbing algorithm, and comparing the optimizing capability.
Experiment 3: aiming at a tabu search algorithm, the influence of the size change of a tabu table and the number change of neighbors searched each time on the average value of the optimal solution is analyzed.
Experiment 4: and comparing the optimizing effects of the tabu search algorithm and the genetic algorithm embedded in the hill climbing algorithm.
In the experiment, a small-scale problem example is used for comparing a tabu search algorithm with a genetic algorithm embedded in a hill climbing algorithm, and relevant parameters of the experiment are designed; the number of market segments I is set to 2, the number of product variants J is set to 2, and the scaling factor μ is set to 0.1. The algorithm-related parameters are designed as:
(1) a genetic algorithm; the population size was set to 300, the crossover rate was set to 0.85, the mutation rate was set to 0.02, and the number of iterations was set to 200.
(2) Tabu search algorithm: the size of the tabu table is set to 11, the number of neighbors in each search is set to 50, and the number of iterations is set to 200.
The results of the numerical experiments are shown in fig. 5 and 6. From observation of the results it can be seen that: the optimizing capability of the genetic algorithm embedded in the hill climbing algorithm is far better than that of the common genetic algorithm in the same running time. Meanwhile, the optimizing capability of the genetic algorithm is far better than that of a tabu search algorithm in the same running time.