WO2018111076A1 - Probabilistic bayesian algorithms for identifying product demand in a small business - Google Patents

Abstract

The invention relates to the use of algorithms for prediction using probabilistic Bayesian networks in order to predict the market demand and the preferences of customers. With the obtained information, the aim is to standardise the small businesses, including financial services which facilitate the consumer to carry out different transactions normally done in supermarkets or chains of convenience stores, without losing sight of the quality of service and attention of any small business.

Description

 PROBABILISTIC BAYESIAN ALGORITHMS FOR THE IDENTIFICATION OF THE DEMAND FOR PRODUCTS IN A SMALL TRADE.

TECHNICAL FIELD OF THE INVENTION

The present invention has its preponderant field of application in the commercial field, more specifically in predicting seasonal demand for a small business in order to offer services comparable to those provided by large commercial chains, without losing sight of the warmth and attention that a small business gives its customers.

BACKGROUND OF THE INVENTION

Classification is an activity that consists of assigning an object to a class or category, the human being performs that task in a natural way to abstract information, leading to a more appropriate representation for decision making. In the case of businesses it is important to have information related to customers and the products they consume, as well as extra factors such as geographical location and preferences. Below are inventions that present methods, processes or systems related to this activity and focused on commercial environments:

The invention US20120303411 presents a system, a method and a computer program for modeling and predicting demand in retail categories. The method uses time series data comprising unit prices and unit sales for a given set of products, with time series data obtained in a given sequence of sales reporting periods and on a collection of stores in a geography of market. Other relevant data sets of participating retail entities that include additional product attribute data, such as market and consumption factors that affect retail demand, are used. A demand model to improve accuracy is achieved through individual steps of the under-modeling method when estimating a model for movements and price dynamics from the time series data of unit prices. In the invention US20140289011 the methods and apparatus implemented by computer for generating prediction markets are presented, the way of calibrating the commercial uncertainties surrounding a project with an uncertain calendar and / or an uncertain result is described. These prediction markets can be used in any segment of the industry and in all business functions, including research and development (R&D), marketing, executive functions and others. Traditional prediction markets, such as equity markets, require liquidity for success. By introducing a pari-mutuel prediction input platform, the present invention describes a modified prediction market that causes more accurate predictions surrounding business decisions.

 The invention US20090254475 describes a method and an apparatus for making predictions for a market, which includes unconventional forecasting options with market participants to determine a prediction framework in which conditional scenarios are found. The method and apparatus contemplates calculating probabilities of realization for each of the conditional scenarios using an approximation calculation technique through an interface, receiving a plurality of predictions associated with selected conditional scenarios, each prediction having an associated value and constructing The market based on predictions. The method and the apparatus comprise updating the probabilities of realization for each of the conditional scenarios in the prediction framework using the approximation calculation technique and establishing the predictions based at least on the updated realization probabilities.

In the invention US20090083128 systems, methods and computer readable media are shown that help assess the probability of success of a new commercial location. Information about existing business locations, including information on a predicted variable, can be provided. Data can be collected from third-party providers, publicly available information or the user who will represent the evaluation variables. A formula is generated that includes evaluation variables and associated coefficients. The coefficients are determined based on a correlation between the evaluation variables and the predicted variable. The data is collected for a new location or business region to determine the value of the evaluation variables of the new location or business region. Applying the coefficients to the values of the evaluation variable for the new location or business region, an output value of the predicted variable is provided. The output value of the predicted variable can be used to assess the probability of success of the new location or business region.

 The invention JP2015032034 provide a demand prediction device, a control method and a program capable of efficiently performing demand prediction of products that have a record. The record related to a similar product of a predicted product (new product) is collected and demand prediction can be done so that the predicted product is before sale, is in an initial state of sale , or has already been sold and a predetermined period has passed since the beginning of the sale according to each period of sale time. In the invention US20070244589 a demand prediction apparatus connected to a record storage unit is presented. The demand prediction apparatus obtains a demand prediction function that conforms to the order reception register, using the acquired order reception register. Next, the demand prediction apparatus calculates a predicted demand value for the product for which the demand prediction is performed, using the derived demand prediction function and issues it.

In the invention US20040260600 a system and method for determining and identifying the demand for articles based on the behavior of trend observation within a member population, such as an online community, is shown. Trends are determined by studying the historical adoption behavior of a group within the member population. The invention US20050197954 provides a system and method for predicting the behavior of small businesses by analyzing data from consumer payment card transactions. The analysis of the speed of transactions and the amount of industry categories and / or profiling based on real-time transactions is used to identify those consumer payment card accounts that are being used improperly to make purchases of small businesses. A small business behavior prediction model is used to record transaction data and update cardholder profiles according to the probability that the transaction data represents the activity of the small business.

DETAILED DESCRIPTION OF THE INVENTION

The characteristic details of the present invention are clearly shown in the following description and in the accompanying figures, which are mentioned by way of example, and therefore should not be considered as a limitation on said invention.

Brief description of the figures:

Figure 1 shows the graph of the client profiling system;

 Figure 2 shows the scheme of the Bayesian network referring to the geographical area of the tooth;

 Figure 3 shows the scheme of the Bayesian network referring to the preference of products and / or services;

Figure 4 shows the scheme of the Bayesian network regarding the classification of products and / or services;

With respect to the figures listed above, Figure 1 shows the elements considered in the client's profiling. The information is obtained by the store clerk directly from the customer and through three agents identifies and classifies geography, preferences and products. This Information is received by the store manager to feed the profiling system. Figure 2 shows the variables to consider in the geographical area of the client: proximity, type of purchase, economy and selected products. Figure 3 shows the variables to consider in the estimation of the preference of products and / or services: quality, price, offers, convenience and intention. The variables to analyze and classify the product and / or service purchased are shown in Figure 4, basically they are main variables (tangibility, customer age, consumption and purchase effort) that are subdivided into factors specific to each variable.

Claims

1. A procedure consisting of prediction algorithms that, through probabilistic Bayesian networks, make predictions related to market demand by seasons, as well as customer preferences in order to standardize small businesses, characterized by:  • Smart agents for classification of information from small businesses related to the demand for items in relation to the season of the year;  • Simple Bayesian functional model for classification by maximizing the argument of the function that describes the probability of belonging to a class, under the assumption of conditional independence between their characteristics;  • Symbolic representation by acyclic graphs of Bayesian networks of conditional independence; 2. The system according to claim No. 1, wherein the classification consists in assigning an object to a class or category in order to extract information to represent it in the most appropriate way for decision making; 3. The system according to claim No. 2, wherein the classification process consists in assigning a class, c, of a set of classes, C, to a certain object or instance, represented by a vector of variables or attributes,

 4. The system according to claim No. 3, wherein there are two basic types of classifiers: Unsupervised or crushing (in this classification, the classes are unknown, and the problem is to divide a set of n objects into / celases, in such a way that similar objects are assigned the same class) and Supervised (in this classification, the classes are known a priori, and the problem is to find a function that assigns each object its corresponding class); 5. The system according to claim No. 4, wherein in the case of small businesses with a commercial turn of the grocery store the classifier the most appropriate is the supervised, since it is known what are the classes to which each product or service belongs; 6. The system according to claim No. 1, wherein the representative functional model uses a mathematical function to map the attributes of the object to its corresponding class:

7. The system according to claim No. 6, wherein the mapping obtained from the model constitutes a set of data D of n elements, each in turn composed of a vector of variables and the corresponding class:

8. The system according to claim No. 7, wherein the criteria for evaluating a classifier are: accuracy (proportion of correct classifications), speed (time it takes to perform the classification), clarity (how understandable it is to humans) and learning time (time to train or adjust the classifier from data);  The system according to claim No. 8, wherein the Bayesian probabilistic approach provides a formal framework for constructing optimal classifiers under certain criteria, such as minimizing the classification error or the cost of a poor classification; 10. The system according to claim No. 9, wherein the Bayesian approach to the supervised classification problem consists in assigning an object described by a set of characteristics, one of

the m possible classes:

such that the probability of the class is obtained by maximizing the function

11. The system according to claim No. 10, wherein the formulation of the Bayesian classifier is based on using the Bayes rule to calculate the subsequent probability of the class, whose mathematical expression is given by:

therefore, the classification problem is expressed as follows:

 Consider a constant and the expression to maximize for the probability of the class is:

12. The system according to claim No. 11, wherein the solution of the classification problem with the Bayesian approach requires an a priori probability for each class P (C), as well as the characteristics of the class P (X \ C) known as plausibility and obtain the subsequent probability

13. The system according to claim No. 12, wherein each characteristic is conditionally independent of the other characteristics given a class, ie

So you have the following equation:

 it is considered a normalization constant; 14. The system according to claim No. 1, wherein the Bayesian networks are based on a semantics of conditional independence between triples of variables by means of a factorization of the joint probability function defined on the n-dimensional random variable;  15. The system according to claim No. 14, wherein the Bayesian model or probabilistic model in a directed acyclic graph is a probabilistic graphic model that represents a set of random variables and their conditional dependencies;  16. The system according to claim No. 15, wherein Bayesian networks are used to represent the profiling of customers, the geographical area of the customer, the preference and classification of products and / or services; 17. The system according to claim No. 16, wherein the knowledge of the relationships between the different variables and their conditional graphic representation allows an estimation of the demand for products in a small business.