KR102577826B1 - Methods and apparatus for generating a predictive model - Google Patents

Abstract

The present invention relates to a method and device for generating a prediction model. The present invention includes a step where the differentiation index calculation unit calculates the new product differentiation index (PDI) by calculating the consumer satisfaction coefficient for each attribute of the new product and the degree of differentiation of each attribute, and the demand creation index calculation unit calculates the sales volume generated by the new product compared to the existing product. A step of calculating the demand creation index (DCI) through relative ratios, and a step of building a demand forecast model through Gaussian process regression of the new product differentiation index (PDI) and demand creation index (DCI) calculated by the demand forecast model construction department. Includes.

Description

Methods and apparatus for generating a predictive model}

The present invention relates to a method and device for generating a prediction model. In particular, it develops a highly complete prediction model that reflects consumer data, and dramatically solves the expert-dependent development process pointed out in existing new product development methodologies to increase the reliability and objectivity of the prediction model. It relates to methods and devices for generating predictive models.

Because it is difficult for most companies to understand the characteristics of the rapidly changing market in advance, there are limitations in implementing effective new product development strategies that meet consumer needs. Even the structure is driven by the opinions of some experts within the company and often relies on the subjective experiences of a small number of people, making it difficult to construct a profile of a new product that maximizes sales volume with objective indicators.

Companies recognize the need for a sales forecasting tool that serves as a standard for evaluating new product development, but because there is no tool that meets this requirement yet, there are limitations in identifying and proposing new product profiles that maximize sales.

Prior patents include Registered Patent No. 10-2289398 (customized big data analysis-based design aesthetics evaluation system for new product development), but the new product development process is structured to judge performance based on demand after product launch. .

The problem that the present invention aims to solve is to solve the problems of the prior art as described above. To predict the sales volume of a new product with new functions added, and to determine the function that can best increase sales performance among the candidate functions of the next new product. The purpose is to provide a method and device for generating a prediction model that can be used for screening.

The prediction model generation method of the present invention includes the steps of calculating the new product differentiation index (PDI) by calculating the consumer satisfaction coefficient for each attribute of a new product and the degree of differentiation of each attribute, and the relative ratio of sales volume generated by the new product compared to existing products. It includes the step of calculating the demand creation index (DCI) and the step of building a demand forecast model through Gaussian process regression of the calculated new product differentiation index (PDI) and demand creation index (DCI).

The prediction model generation device of the present invention includes a differentiation index calculation unit that calculates how differentiated the attributes included in the new product are from existing products, a demand creation index calculation unit that calculates the ratio of the difference in initial sales volume between the existing product and the new product, and It consists of a demand forecast modeling construction department that builds the new product differentiation index (PDI) and demand creation index (DCI) through Gaussian process regression.

The present invention can effectively improve the new product development process by breaking away from the product development method that relies heavily on subjective attributes and presenting an objective prediction model generation method using a data scientific approach.

The present invention can fully reflect market needs in model learning for products that will maximize future demand creation in the very early stages of new product development by reflecting market reactions in a prediction model.

Figure 1 is a flowchart illustrating a method of building a demand forecast model in the forecast model generation method according to an embodiment of the present invention.
Figure 2 is a configuration diagram of a predictive new product development device according to an embodiment of the present invention.
Figure 3 is a table showing the consumer satisfaction coefficient calculated by the Kano survey according to an embodiment of the present invention.
Figure 4 is a table showing an example of calculating the differentiation index for each attribute of a new product according to an embodiment of the present invention.
Figure 5 is a table showing normalized values compared to the differentiation index according to an embodiment of the present invention.
Figure 6 is a graph showing the calculated value of the demand creation index according to an embodiment of the present invention.
Figure 7 is a table showing the results of the Gaussian process regression prediction model according to an embodiment of the present invention.
Figure 8 is a table showing accuracy through cross-validation according to an embodiment of the present invention.
Figure 9 is a table showing predicted sales volume for each new product scenario including new attributes according to an embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are merely illustrative for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention are It may be implemented in various forms and is not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention can make various changes and have various forms, the embodiments will be illustrated in the drawings and described in detail in this specification. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes all changes, equivalents, or substitutes included in the spirit and technical scope of the present invention.

The terms used in this specification are merely used to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in this specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings attached to this specification.

1 is a flowchart illustrating a method of building a demand forecast model in the forecast model generation method according to an embodiment of the present invention. Referring to Figure 1, the prediction model generation method first calculates the new product differentiation index (PDI) by calculating the consumer satisfaction coefficient and the degree of differentiation in the differentiation index calculation unit 110 (S101).

The New Product Differentiation Index (PDI) measures how differentiated the attributes included in a new product are compared to existing products (Predecessor products). To calculate the new product differentiation index (PDI), the consumer satisfaction coefficient and the degree of differentiation for each attribute are used as weights.

First, in order to obtain the consumer satisfaction coefficient, the consumer satisfaction coefficient calculation module 111 uses a method that can precisely identify consumer requirements, such as the KANO model, to reflect the importance of each attribute from the consumer's perspective. The KANO model can systematically explain consumers' reactions to products by expressing them as a dual perception of satisfaction and dissatisfaction, and is widely used in the field of new product development to analyze consumer reactions for each product attribute. . The KANO model evaluates each attribute of a product using six quality factors, as shown in <Table 1> below, according to the impact of the satisfaction of each attribute on the consumer's satisfaction.

quality factors explanation Attractive Quality Element(A) A type that satisfies consumers with features they did not expect from the product. If the consumer is satisfied, satisfaction increases significantly, and if the consumer is not satisfied, it decreases slightly. One-Dimensional Quality Element(O) The most common type of product. If the consumer is satisfied, satisfaction increases, and if the consumer is not satisfied, it decreases. Must-be Quality Element(M) Features that a product must have in order for consumers to be satisfied. If the consumer is not satisfied, satisfaction decreases significantly, and if satisfied, it rises slightly. Indifferent Quality Element(I) A characteristic that consumers are not interested in about a product. Regardless of whether the consumer is satisfied or not, the consumer's sense of satisfaction is constant. Reverse Quality Element(R) A characteristic that is the opposite of a one-dimensional quality factor. If the consumer is satisfied, satisfaction decreases, and if the consumer is not satisfied, satisfaction increases. Skeptical Quality Element(S) Features of a product that consumers have doubts about understanding. Regardless of satisfaction, consumers may or may not be satisfied. Not used to measure satisfaction coefficient.

In order to index the consumer's satisfaction and express it as a consumer satisfaction coefficient, the KANO model classifies the quality factors of each product and then calculates it in the consumer satisfaction coefficient calculation module 111 using the consumer satisfaction coefficient formula below. .

The consumer satisfaction coefficient is derived by calculating the proportion of attractive, unified quality elements in the sum of all quality element responses, and the consumer satisfaction coefficient for each attribute of the product can be determined.

Next, the degree of differentiation calculation module 113 determines the degree of differentiation for each attribute. Product differentiation occurs in the form of feature attributes being improved (or improved) or new attributes added to existing products. In order to accurately calculate the differentiation index, it is necessary to determine the degree of differentiation applied to each attribute of the new product compared to the existing product. To this end, it is necessary to distinguish the degree of differentiation for each function of the existing product and the new product.

In the present invention, attributes that are the same as or deteriorated from existing products are 'non-differentiation', improving the attributes of an existing product by 20% or less is 'slight differentiation', and improving existing attributes by more than 20% or adding completely new attributes is 'non-differentiation'. The case is classified as ‘strong differentiation’. If an attribute is difficult to measure arithmetically, the degree of differentiation is determined based on expert evaluation. For attributes where differentiation has occurred, the weight for the degree of differentiation is 0, 0.2, and 0.7 for no differentiation, weak differentiation, and strong differentiation, respectively.

In order to calculate the differentiation index of the new product, the differentiation index calculation unit 110 reflects the two types of weights discussed above, namely, the consumer satisfaction coefficient for each attribute of the new product and the degree of differentiation of the new product. A more detailed calculation method is as follows.

First, the consumer satisfaction coefficient for each attribute is multiplied by the value of the degree of differentiation of the new product to obtain the differentiation index for each new product attribute ( ) is obtained. Afterwards, the differentiation indices for all attributes are used to calculate comprehensive differentiation indices, which are values that express the differences between new products and existing products in comprehensive numbers, and these are collectively referred to as the Product Differentiation Index (PDI). Below is an equation that expresses this.

Here, n is the number of differentiated attributes in the new product among those selected as core attributes of the product.

The differentiation index (PDI) of the new product requires a normalization process in order to use the differentiation index (PDI) between existing products and new products in modeling with the demand creation index. The comprehensive differentiation index is calculated based on the Euclidean distance between existing and new product attributes. If there is a wide range of values among the values obtained through this process, a problem arises in which the modeling results are greatly influenced by the values, which can affect the performance of the model.

To normalize the differentiation index (PDI), maxPDI is obtained by assuming that a new product has maximum improvement in all key attributes from an existing product. In other words, maxPDI is obtained by calculating PDI assuming that all core properties and functions are strongly differentiated. Conversely, minPDI is obtained assuming a situation where a new product has little improvement in all key attributes and functions from the existing product. In other words, minPDI is obtained by calculating PDI assuming that all core properties and functions are basic. As a result, it can be seen that minPDI is 0 in all situations. Finally, the normalized value z of the differentiation index PDI between existing products and new products can be obtained using PDI, maxPDI, and minPDI.

Next, in order to predict demand through the new product differentiation index (PDI) previously derived in the demand creation index calculation unit 120, the demand creation index (DCI) is calculated through the relative ratio of the sales volume generated by the new product. Calculate (S103).

The demand creation index (DCI) refers to the value expressed by the differentiation index (PDI) of the new product as the ratio of the difference in initial sales volume of the existing product and the new product, and the formula below represents the demand creation index (DCI) of the new product. If the demand creation index (DCI) is 0, it means that the sales volume is the same as that of the existing product. A positive number greater than 0 indicates that the sales volume of the new product is greater than that of the existing product. A negative number less than 0 indicates that the sales volume of the new product is greater than that of the existing product. It indicates little.

Next, the demand forecast model construction unit 130 builds a demand forecast model through Gaussian process regression (S105). Through the above demand forecasting model, the sales volume of new products with new functions can be predicted, and it can be used to select the function that can best increase sales performance among the candidate functions of the next new product.

In order to build the demand forecasting model, an algorithm that can accurately model the relationship between the two is used to predict the demand creation index (DCI) using the previously obtained new product differentiation index (PDI). Through various previous arguments, it has been shown that the relationship between product differentiation and demand is not linear, and in particular, market data related to new products are complex and have stronger non-linear characteristics. In the present invention, the Gaussian process is selected as the optimal model.

The Gaussian process regression model is a non-parametric model that solves non-linear problems and is a representative regression model that generates an advanced prediction model even with small data. Unlike general machine learning models, this is a non-parametric model that relies on a small number of parameters to create a prediction model. The Gaussian process is a technique that generates a Gaussian process latent variable f(Xi) for the observed value Xi and predicts the y value by introducing an explicit basis function h. When expressed as a formula, it takes the form below.

In the above formula, m(X) means the mean and k(X, Its characteristic is to derive and make predictions. Even by learning a small amount of data, numerous operations are performed to calculate each similarity, resulting in higher performance than general prediction models, so the Gaussian process regression model can be effectively used in cases such as predicting new products.

The Gaussian process is derived using the above calculation formula and is combined with the regression equation as follows. At this time, h is an explicit basis function that describes the response variable for each observation.

Finally, Bayes' theorem is used in the formula to derive the optimal y value that reflects the existing results, that is, the final predicted value of DCI. The DCI forecast value derived in this way is multiplied by the demand for the existing product to calculate the final result of the model, that is, the predicted demand for the first quarter after the launch of the new product. This is expressed in the equation as follows.

Figure 2 is a configuration diagram of a prediction model generating device 100 according to an embodiment of the present invention. Referring to FIG. 2, the prediction model generation device 100 includes a differentiation index calculation unit 110, a demand creation index calculation unit 120, a demand forecast model construction unit 130, a storage unit 140, and a communication unit 150. , and consists of a control unit 160.

The differentiation index calculation unit 110 calculates a new product differentiation index (PDI) to measure how differentiated the attributes included in the new product are from existing products. The new product differentiation index (PDI) uses the consumer satisfaction coefficient and the degree of differentiation for each attribute as weights. The differentiation index calculation unit 110 is composed of a consumer satisfaction coefficient calculation module 111 and a differentiation degree calculation module 113. Details of the differentiation index calculation unit 110 will be described later.

The consumer satisfaction coefficient calculation module 111 calculates the consumer satisfaction coefficient using a method that can precisely identify consumer requirements, such as the Kano model, in order to reflect the importance of each attribute from the consumer's perspective. In order to index consumer satisfaction using the KANO model and express it as a consumer satisfaction coefficient, the quality factors of each product are classified and then calculated using the consumer satisfaction coefficient formula.

The differentiation degree calculation module 113 calculates the degree of differentiation for each function of the existing product and the new product in order to determine the degree of differentiation for each attribute. The weights for the degree of differentiation are 0, 0.2, and 0.7 for no differentiation, weak differentiation, and strong differentiation, respectively. The above 'non-differentiation' refers to attributes that are the same as or deteriorated from existing products, the above 'weak differentiation' refers to improving the attributes of existing products by 20% or less, and the above 'strong differentiation' refers to improving the existing attributes by 20% or less. This refers to cases where improvements are made by more than % or completely new properties are added. If an attribute is difficult to measure arithmetically, the degree of differentiation is determined based on expert evaluation.

The differentiation index calculation unit 110 multiplies the consumer satisfaction coefficient for each attribute obtained from the consumer satisfaction coefficient calculation module 111 by the value of the degree of differentiation of the new product obtained from the differentiation degree calculation module 113 to obtain a differentiation index for each new product attribute. ( After calculating, the differentiation indices for all attributes are used to calculate the comprehensive differentiation index, which is a value that expresses the difference between the new product and the existing product in a comprehensive number. Afterwards, since a normalization process is necessary for demand creation index and modeling, PDI, maxPDI, and minPDI are used to calculate the normalized value z of the differentiation index PDI between existing products and new products.

The demand creation index calculation unit 120 calculates the demand creation index (DCI) through the relative ratio of the sales volume generated by the new product in order to predict demand through the previously derived differentiation index (PDI) of the new product.

The demand forecast model building unit 130 predicts the sales volume of new products with new functions by building a demand forecast model through Gaussian process regression, and selects the function that can increase sales performance the most among the candidate functions of the next new product. You can pick it out.

The storage unit 140 stores each component value of a process related to building a demand forecast model.

The communication unit 150 communicates on the network with the prediction model generation device 100, the service providing server, and other user terminals, and uses the results of the KANO model as weights to make highly complete predictions reflecting data through product users' satisfaction with each function. Receive information to study the model. The network includes a local area network (LAN), a wide area network (WAN), a value added network (VAN), a mobile radio communication network, a satellite communication network, and combinations thereof. Includes. It is a data communication network in a comprehensive sense and may include wired Internet, wireless Internet, and mobile wireless communication networks. In addition, wireless communications include, for example, wireless LAN (Wi-Fi), Bluetooth, Bluetooth low energy, ZigBee, WFD (Wi-Fi Direct), UWB (ultra wideband), and infrared communications (IrDA). Data Association), NFC (Near Field Communication), etc., but are not limited to these.

The control unit 160 controls the processing of processes related to building a demand forecast model and controls the operation of each component.

Below is an example of application to a smartwatch new product development case to verify the effectiveness of the predictive new product development method and system of the present invention. Smartwatches are suitable for applying the present invention because, due to the nature of the product, new products are regularly and consistently released in the same product line, and it is a development field in which planning for innovative functions frequently occurs.

In this study, data from Apple Watch and Samsung Electronics' Galaxy Watch, for which it is relatively easy to obtain data and for which there are sufficient comparable new product series, were analyzed. The analyzed products include Apple Watch Series 1~7 and Apple Watch SE, which were new products released in September 2014, and Galaxy Watch Series 1~3, Galaxy Watch Active 1~2, and Galaxy Watch Classic, which were released in August 2018. It has been done. The analysis was conducted based on data from the Korean region. Korea has a high level of maturity in IT technology and is used as a test bed for new IT products, so many studies use data from the Korean market.

First, to calculate PDI, the consumer satisfaction coefficient and the degree of differentiation by product attribute are obtained. A total of 20 key smartwatch attributes were selected based on preliminary interviews with experts and potential users.

Kano survey data is used to obtain consumer satisfaction coefficients for each attribute. Referring to Figure 3, responses to the positive and negative questions of the KANO model were received on a 5-point Likert scale. Based on the corresponding KANO survey data, quality factors and satisfaction coefficients for a total of 20 smartwatch attributes were calculated.

Next, the degree of differentiation for each attribute is identified. To determine the degree of differentiation, a combination of new products and existing products (predecessor products) is needed. Referring to Figure 4, as an example of the level of improvement between existing products and new products carried out in the present invention, the degree of differentiation of some attributes for Apple Watch Series 3 and Apple Watch Series 4 is distinguished. Based on the criteria described above, 0.2 is given for weak differentiation and 0.7 for strong differentiation.

The differentiation index for each attribute can be obtained by multiplying the satisfaction coefficient for each attribute obtained previously by the degree of differentiation. For attributes where differentiation has not occurred, the degree of differentiation becomes 0, so the differentiation index for each attribute also becomes 0, and for attributes where differentiation has occurred, a differentiated 'differentiation index for each attribute' is calculated according to the degree of differentiation and the consumer satisfaction coefficient. Figure 4 is an example of calculating the differentiation index by attribute of a new product for Apple Watch Series 4 compared to Apple Watch 3. Afterwards, the PDI of the new product is obtained by calculating the overall differentiation index using the differentiation index for each new product attribute.

To model the differentiation index and demand creation index of new products, the differentiation index is normalized. When comparing a new smartwatch to an existing smartwatch, maxDI is calculated assuming a situation in which all functions are strongly differentiated, and minDI is calculated assuming a situation in which no differentiation occurs at all. Normalize the differentiation index of new products using maxDI and minDI. Figure 5 is an example of calculating the normalized value (z) compared to the differentiation index of Apple Watch 4 compared to Apple Watch 3.

To calculate the percentage change in sales of new smartwatches compared to existing smartwatches, or DCI, smartwatch sales data is used to list the initial demand for existing and new products in pairs.

DCI uses the initial demand for a new product divided by the initial demand for an existing product within each observation location to measure the rate of change in sales volume. This becomes the dependent variable for model learning. Figure 6 is an example of calculating the DCI value of smartwatch data released by Apple.

Once the PDI and DCI of a new smartwatch are secured, work is done to build a prediction model by fitting the relationship between the two indices. For high-quality predictive modeling, 23 complete data points excluding outliers are selected during the data preprocessing process, 80% of which are used as training data, and the remaining 20% are used as test data. At this time, the test data is randomly extracted, but set so that it does not predict the same new product, so that it can predict as many different smartwatches as possible.

After completing the data preprocessing process for model learning, a prediction model of Gaussian process regression is created. In this process, Gaussian process regression can be predicted for the 20% of test data selected previously by drawing a prediction trend line as shown in Figure 7.

Referring to Figure 7, you can see the results of the Gaussian process regression prediction model proposed in this study using the development of a new smartwatch as an example. This study's DCI prediction model using the Gaussian process regression technique shows that the accuracy of DCI prediction is 0.097 based on MSE, and the overall mean square error shows high accuracy. The closer the DCI forecast is to the actual value, the more it has a direct impact on the initial sales volume forecast for a new product. Therefore, the final result of the prediction model, that is, the initial sales volume prediction performance, also recorded 9.6% based on MAPE, confirming relatively high prediction performance.

The robustness of the model results was confirmed through cross validation. We repeated the test by randomizing an additional subset of the entire dataset of the same size as the baseline test. Referring to Figure 8, as a result of 5-folds CV, the average performance of all tests was MSE 0.205 and MAPE 12.2%, maintaining a relatively high level of accuracy.

Based on the prediction model built in this example, sales volume for each new product scenario with new attributes can be predicted. Referring to Figure 9, you can see the sales volume forecast for the first quarter after the launch of each new product profile and the sales contribution of the new attribute. The sales contribution of new attributes was calculated through the difference in predicted sales volume when including and not including the attribute in a Gaussian process regression model, which indicates the impact of each new attribute on demand.

The invention has been described with reference to the embodiments shown in the drawings, but these are merely illustrative, and those skilled in the art will understand that various modifications and other equivalent embodiments are possible therefrom. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the attached registration claims.

100; Predictive model generating device 110; Differentiation Index Calculation Department
111; Consumer satisfaction coefficient calculation module 113; Differentiation degree calculation module
120; Demand Creation Index Calculation Department 130; Demand Forecast Model Construction Department
140; storage unit 150; Ministry of Communications
160; control unit

Claims (5)

A differentiation index calculation unit calculates a new product differentiation index (PDI) by calculating the consumer satisfaction coefficient for each attribute of the new product and the degree of differentiation of each attribute;
A step where the demand creation index calculation unit calculates the demand creation index (DCI) through the relative ratio of sales volume generated by new products compared to existing products; and
It includes a step where the demand forecast model construction unit builds a demand forecast model through Gaussian process regression of the calculated new product differentiation index (PDI) and demand creation index (DCI),
The New Product Differentiation Index (PDI) is a value that expresses the difference between a new product and an existing product as a comprehensive value,
The new product differentiation index (PDI) is calculated by multiplying the consumer satisfaction coefficient for each attribute by the value of the degree of differentiation of the new product and calculating the differentiation index for each new product attribute ( ) is calculated using the equation below using the differentiation indices for all attributes,
The degree of differentiation of a new product is calculated by calculating the degree of differentiation for each function of the existing product and the new product. The weight for the degree of differentiation is 0 for no differentiation, 0.2 for weak differentiation, and 0.7 for strong differentiation.
The consumer satisfaction coefficient includes the steps of classifying the quality factors of each product using the KANO model; and
A method of creating a prediction model that includes steps calculated by applying the consumer satisfaction coefficient formula.

(i=1,…,n, where n is the number of differentiated attributes in the new product among those selected as core attributes of the product)

delete delete According to paragraph 1,
The demand creation index (DCI) refers to the value expressed by the new product differentiation index (PDI) as the ratio of the difference in initial sales volume of existing products and new products, and is a prediction model generation method calculated by the formula below.
A differentiation index calculation unit that calculates how different the attributes included in a new product are compared to existing products;
A demand creation index calculation unit that calculates the ratio of the difference in initial sales volume of existing products and new products; and
It consists of a demand forecast modeling construction department that builds the new product differentiation index (PDI) and demand creation index (DCI) through Gaussian process regression.
The new product differentiation index (PDI) is a value that expresses the difference between the new product and the existing product as a comprehensive value,
The differentiation index calculation unit calculates the degree of differentiation of each function of the existing product and the new product as a numerical value of the degree of differentiation of the new product, and the weight for the degree of differentiation is assigned 0 to no differentiation, 0.2 to weak differentiation, and 0.7 to strong differentiation. A prediction model generating device characterized in that:

Images