KR102609726B1 - System for recommending personalized fruit and vegetables based on artificial intelligence - Google Patents

Abstract

An artificial intelligence-based personalized fruit and vegetable recommendation system is launched. The artificial intelligence-based personalized fruit and vegetable recommendation system collects the user's lifelog, provides it to a cloud server, and selects personalized fruit and vegetables based on personalized fruit and vegetable-gene activity information obtained from the cloud server. user terminal; and a cloud server that generates personalized fruit-vegetable-gene activity information by linking with a pre-built fruit and vegetable database, a nutrient database, and an external server.

Description

Personalized fruit and vegetable recommendation system based on artificial intelligence {SYSTEM FOR RECOMMENDING PERSONALIZED FRUIT AND VEGETABLES BASED ON ARTIFICIAL INTELLIGENCE}

The present invention relates to fruit and vegetable recommendation technology, and more specifically, to a personalized fruit and vegetable recommendation system based on artificial intelligence.

As dietary culture has become more sophisticated and people's interest in and awareness of health have increased significantly, the needs of people who want to eat healthy food are increasing significantly.

In particular, fruits and vegetables such as fruits, vegetables, and greens are natural foods, and at the same time, they are foods that can be consumed in a balanced manner with various vitamins and nutrients, and they also taste good, so many people consume them and look for them in various ways.

However, not only do these fruits and vegetables mainly grow in different seasons, times, and places for each fruit and vegetable, but the various nutrients, including the unique vitamins, are all slightly different for each fruit and vegetable, making it difficult for consumers to consume fruits and vegetables with the desired nutrients at the appropriate time. There is a difficult problem.

As a way to solve this problem, research is being conducted on customized food recommendation technology that selects foods containing widely known nutrients based on simple information such as an individual's physical characteristics or disease.

However, in the case of conventional customized food recommendation technology, most of the matching nutrients and contents are built in a simple database based on the individual's simple physical characteristics or diseases and then select matching foods, so each individual has different physical characteristics. When there are characteristics and genetic predispositions, it is difficult to take them all into consideration.

Meanwhile, interest in and research on machine learning technology is increasing recently, and research on artificial neural networks based on deep learning, a field of machine learning, is active.

These artificial neural networks generally operate by classifying classes of input data based on pre-provided learning data, but artificial neural networks can vary significantly depending on the composition and learning method of data to properly operate and learn the artificial neural network. Because there are differences in the classification performance of neural networks, the developer's design capabilities are more important than anything else.

However, the conventional customized food recommendation technology is only at the level of simple use of a widely known artificial neural network, and there is a problem in that learning data and input data that match the data type and structure of the artificial neural network must be individually collected and utilized. There are many difficulties in learning data and using it as input data to achieve the desired level of performance.

The purpose of the present invention to solve the above problems is to provide a personalized fruit and vegetable recommendation system based on artificial intelligence.

To achieve the above object, the present invention provides a personalized fruit and vegetable recommendation system based on artificial intelligence.

The artificial intelligence-based personalized fruit and vegetable recommendation system collects the user's lifelog, provides it to a cloud server, and selects personalized fruit and vegetables based on personalized fruit and vegetable-gene activity information obtained from the cloud server. user terminal; and a cloud server that generates personalized fruit-vegetable-gene activity information by linking with a pre-built fruit and vegetable database, a nutrient database, and an external server.

The user terminal generates a plurality of input vectors using the fruit and vegetable-gene activity information, sequentially inputs the plurality of generated input vectors into a supervised learning artificial neural network, and Based on the output values, a number of fruits and vegetables are selected to recommend to the user.

The fruit and vegetable database stores content information indicating the content of various nutrients contained in each fruit and vegetable.

The nutrient database includes metabolic amount information indicating the amount of nutrients each nutrient is metabolized by the body's metabolic process and absorbed into the body; and activity information indicating the degree to which each metabolite inhibits or activates gene mutation or gene expression when absorbed into the body.

The cloud server acquires medical information in conjunction with the external server, determines weight values for each gene corresponding to the user based on the medical information and the life log, and generates a gene weight indicating the determined weight values. generate information.

The cloud server generates a first matrix using the content information, generates a second matrix using the metabolic amount information, generates a third matrix using the activity information, and uses the gene weight information. A fourth matrix is generated, and a fifth matrix corresponding to the fruit and vegetable-gene activity information is obtained through a matrix multiplication operation between the first matrix and the fourth matrix.

The artificial neural network is supervised in advance using training data collected in advance.

The training data uses input vectors obtained from the rows of the fifth matrix corresponding to each of the fruit-gene activity information obtained for a number of users other than the user as training input values for the artificial neural network, and It is a set of data in which the target output vector corresponding to the fruits and vegetables most purchased or viewed by a large number of other users is used as a training output value for the artificial neural network.

The user terminal groups the selected plurality of fruits and vegetables into sets of fruits and vegetables ranging from 2 to n (n is the total number of the selected plurality of fruits and vegetables), and between the fruits and vegetables belonging to each of the sets The compatibility degrees are calculated for each of the sets, and among the sets for which the calculated compatibility degree is higher than a preset threshold, the fruits and vegetables in the set selected according to the preset criteria are provided to the user as fruits and vegetables for the subscription service.

When using the artificial intelligence-based personalized fruit and vegetable recommendation system according to the present invention as described above, not only the user's life log, but also the nutrient content of each fruit and vegetable, the amount of metabolism of metabolites according to nutrients, gene activity due to absorption of metabolites, and personal It is possible to recommend fruits and vegetables optimized for an individual by comprehensively considering whether or not they possess specific genes.

In addition, the learning performance of the artificial neural network is improved by setting the target output vector corresponding to the fruit and vegetable according to the unique characteristics of the fruit and vegetable, and the loss function is defined by considering all target output vectors corresponding to each fruit and vegetable, thereby improving the artificial neural network according to supervised learning. It has the advantage of further improving the output accuracy of the neural network.

Figure 1 is a conceptual diagram of an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment.
FIG. 2 is a block diagram showing functional components of the cloud server according to FIG. 1.
FIG. 3 is a conceptual diagram for explaining the operation of the fruit and vegetable-gene activity information generation unit according to FIG. 2.
FIG. 4 is a block diagram showing functional components of the user terminal according to FIG. 1.
FIG. 5 is a diagram showing the structure of an artificial neural network linked to the user terminal according to FIG. 1.
Figure 6 is a first diagram showing an application operating on a user terminal by an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment.
Figure 7 is a second diagram showing an application operating on a user terminal by an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment.
FIG. 8 is a diagram illustrating an exemplary hardware configuration of the user terminal according to FIG. 1.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present invention. While describing each drawing, similar reference numerals are used for similar components.

Terms such as first, second, A, and B may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be named a second component, and similarly, the second component may also be named a first component without departing from the scope of the present invention. The term and/or includes any of a plurality of related stated items or a combination of a plurality of related stated items.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the 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.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless explicitly defined in the present application, should not be interpreted in an ideal or excessively formal sense. No.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings.

Figure 1 is a conceptual diagram of an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment.

Referring to FIG. 1, the artificial intelligence-based personalized fruit and vegetable recommendation system 1000 according to an embodiment collects the user's lifelog and provides it to the cloud server 300, and collects the user's lifelog from the cloud server 300. A user terminal 100 that selects personalized fruit and vegetables based on the acquired personalized fruit and vegetable-gene activity information, and a personalized fruit and vegetable database 310 linked with a pre-built fruit and vegetable database 310 and a nutrient database 320 and an external server 400. It may include a cloud server 300 that generates fruit and vegetable-gene activity information.

The user terminal 100 may collect the user's lifelog in conjunction with at least one wearable device 200. Here, lifelog refers to indicators representing various types of lifestyle habits or living conditions that can be collected during the user's daily life and may include, for example, heart rate, body temperature, sleep time, number of steps, heart rate, and weight. You can.

Examples of the user terminal 100 include a desktop computer, laptop computer, laptop, smart phone, tablet PC, and mobile phone capable of communication. ), smart watch, smart glass, e-book reader, PMP (portable multimedia player), portable game console, navigation device, digital camera, DMB (digital multimedia broadcasting) It may be a player, a digital audio recorder, a digital audio player, a digital video recorder, a digital video player, a PDA (Personal Digital Assistant), etc.

Additionally, the user terminal 100 may collect lifelogs by receiving input from the user. For example, the lifelog input from the user may include the type and amount of food consumed during the day, the time at which the meal was eaten, etc.

The user terminal 100 may receive personal identification information from the user. Here, personal identification information may include the user's age, gender, region of birth, resident registration number, etc. Personal identification information may be de-identified and stored in the user terminal 100.

The user terminal 100 may transmit the lifelog and personal identification information collected here to the cloud server 300.

The cloud server 300 can generate personalized fruit and vegetable-gene activity information for the user based on the life log and personal identification information. Here, the fruit and vegetable-gene activity information may be data that uniquely represents the correlation between a number of fruits and vegetables and a number of genes for the user.

In one embodiment, fruits and vegetables may be interpreted as including not only fruits such as apples, tangerines, oranges, and grapes, but also vegetables such as onions, garlic, lettuce, and spinach.

In one embodiment, a gene may be defined as specific genomic sequences constituting a chromosome, or may be interpreted to mean a protein expressed by such genomic sequences.

In particular, in one embodiment, the genes may include genes known to be associated with various types of diseases such as diabetes or high blood pressure.

For example, in the case of diabetes, mutations in the PAX4 (paired box 4) gene, which is involved in the differentiation of pancreatic beta cells that secrete insulin, and mutations in the GLP1R (glucagon-like peptide 1 receptor) gene, an incretin hormone receptor used as a diabetes injection treatment, Mutations are known to have a major impact on disease development.

For example, if there is a mutation in the PAX4 gene, the age of onset of diabetes is lowered and the risk of diabetes increases by about 1.5 times, and it is known to occur specifically in Asians such as Koreans.

In addition, if there is a mutation in the GLP1R gene, the risk of diabetes is reduced by 0.86 times, and the cardiovascular risk of diabetic patients is reduced, and it is known to occur specifically in Asians.

In the case of hypertension as well as diabetes, it is known that the risk of hypertension varies depending on mutations in the EML6 gene, MYL2 gene, and JAG1 gene.

For example, among the group that consumed less than 2g of sodium per day, the risk of hypertension was found to be 1.2 times higher if there was an EML6 gene mutation, and in the group with mutations in the MYL2 gene and JAG1 gene, the risk of hypertension was lower than that of the general population even if salt intake was high. It is said that it was 0.8 times lower than that.

Therefore, in one embodiment, genes can be interpreted as including all genes that affect physical characteristics, such as various types of diseases or pupil color or hair color, including the above-mentioned PAX4, GLP1R, and EML6 genes.

The cloud server 300 may operate in conjunction with the fruit and vegetable DB 310 and the nutrient DB 320 to generate personalized fruit and vegetable-gene activity information. The fruit and vegetable DB 310 and the nutrient DB 320 may be implemented in the form of a distributed database connected to the cloud server 300 via a wired or wireless network, or may be implemented in the form of a database using the internal storage of the cloud server 300. .

Specifically, the fruit and vegetable DB 310 and the nutrient DB 320 use each of a plurality of physical computing devices as storage to physically distribute and store data, but logically store the data in the form of a single storage cloud server 300. It can be managed by . For example, the fruit and vegetable DB 310 and the nutrient DB 320 are Java-based large data distributed processing systems that store data in HDFS (Hadoop Distributed File System) and process data using MapReduce. It can be implemented with Hadoop.

The fruit and vegetable DB 310 can store information about various nutrients contained in various fruits and vegetables. Specifically, the fruit and vegetable DB 310 can store content information indicating the content of various nutrients contained in each fruit and vegetable. Here, nutrients may refer to various minerals including vitamin C, calcium, potassium, iron, fatty acids, etc. contained in fruits and vegetables.

The nutrient DB 320 may store information representing the association between multiple nutrients and multiple metabolites and information representing the association between metabolites and multiple genes. Here, metabolites are substances obtained through the body's metabolism when nutrients are absorbed into the body, and may be amino acids, cholesterol, alkaloids, pigments, etc.

Specifically, the nutrient DB 320 may store metabolic amount information indicating the amount of nutrients each nutrient is metabolized by the body's metabolic process and absorbed into the body. Additionally, the nutrient DB 320 may store activity information indicating the degree to which each metabolite suppresses or activates gene mutation or gene expression when absorbed into the body.

At this time, at least some of the target substances may overlap with nutrients. For example, at least some of the nutrients may be absorbed without change through body metabolism, and in this case, they may be treated as metabolites and nutrients at the same time.

Additionally, the cloud server 300 can acquire medical information by linking with the external server 400. For example, the cloud server 300 may obtain medical information corresponding to the user from the external server 400 using personal identification information obtained from the user terminal 100. Medical information may include medical records, health examination records, or genetic test records of the user receiving treatment at a medical institution.

The cloud server 300 may determine weight values for each of the genes corresponding to the user based on the medical information and the lifelog obtained from the user terminal 100 and generate gene weight information representing the determined weight values. there is.

In other words, gene weight information is either possessed by the user (through genetic test records) or presumed to be possessed by the user (through medical records or health check-up records, etc.), and is used to increase the likelihood of expression of the gene according to the life log. If there is an effect, the weight value corresponding to the corresponding gene can be set high, and the low weight value can be set low for genes that the user does not have or are estimated to not have.

To generate such genetic weight information, various types of statistical analysis methods or probability analysis methods can be used based on various medical records, life logs, health examination records, etc. For example, based on the user's medical record, people with similar medical records or medical history are searched, and the greater the number of people with the gene in the genetic test records of the searched people, the weight of the user corresponding to the gene is increased. The value can be set high.

In other words, gene weight information is used as information representing the weight value of genes that are uniquely expressed, possessed, or mutated for each individual, and is information that can recommend the most suitable fruits and vegetables at the level of genes fundamentally possessed by each person. It is used as.

The cloud server 300 can generate personalized fruit-gene activity information for the user by comprehensively using content information, metabolism information, activity information, and gene weight information.

The cloud server 300 may provide the generated fruit and vegetable-gene activity information to the user terminal 100.

The user terminal 100 generates a plurality of input vectors using fruit-gene activity information, sequentially inputs the generated plurality of input vectors into the supervised learning artificial neural network 10, and artificially Based on the output value of the neural network 10, fruits and vegetables to recommend to the user can be selected. The user terminal 100 may display the selected fruits and vegetables to the user.

At this time, the user terminal 100 may select a plurality of fruits and vegetables to recommend to the user and calculate the degree of compatibility between the selected fruits and vegetables. For example, the user terminal 100 groups a plurality of fruits and vegetables into sets with a number between 2 and n (where n is the total number of selected fruits and vegetables), and sets a number between the fruits and vegetables within each set. Compatibility levels are calculated, and among a set of fruits and vegetables whose calculated compatibility level is greater than or equal to a preset threshold, fruits and vegetables in the set selected according to preset criteria can be provided to the user as fruits and vegetables for a subscription service.

Here, the subscription service is a service that provides automatic purchase of a plurality of fruits and vegetables sequentially every week or every month. Among the users of the user terminal 100, the user terminal 100 of the user who has subscribed to this subscription service, Fruits and vegetables for the subscription-type service can be automatically ordered by linking with the cloud server 300 at each pre-designated subscription period.

To this end, the cloud server 300 may communicate with the user terminal 100 to perform procedures for ordering, payment, and delivery of pre-registered fruits and vegetables. For example, the cloud server 300 receives an order request from the user terminal 100, transmits a payment request to the user terminal 100, receives payment approval from the user terminal 100, and sends an order request and A delivery request for the corresponding fruits and vegetables can be transmitted to the operation server of a delivery company that provides an external delivery service.

FIG. 2 is a block diagram showing functional components of the cloud server according to FIG. 1. FIG. 3 is a conceptual diagram for explaining the operation of the fruit and vegetable-gene activity information generation unit according to FIG. 2.

Referring to FIG. 2, the cloud server 300 includes a personal information acquisition unit 301 that acquires life logs and personal identification information in conjunction with the user terminal 100; A DB management unit (302) that acquires content information, metabolic rate information, and activity information in conjunction with the fruit and vegetable DB (310) and the nutrient DB (320); a gene weight information generator 303 that generates gene weight information corresponding to the user using medical information and life logs obtained in conjunction with the external server 400; and a fruit-vegetable-gene activity information generation unit 304 that generates fruit-vegetable-gene activity information using content information, metabolic amount information, activity information, and gene weight information.

The DB management unit 302 may access the fruit and vegetable DB 310 to obtain content information and access the nutrient DB 320 to obtain metabolic information and activity information.

The genetic weight information generator 303 acquires medical information in conjunction with the external server 400, and generates each of the genes corresponding to the user based on the obtained medical information and the lifelog obtained from the user terminal 100. Weight values can be determined, and gene weight information representing the determined weight values can be generated.

The fruit and vegetable-gene activity information generation unit 304 generates a first matrix using content information, a second matrix using metabolic information, a third matrix using activity information, and gene weights. A fourth matrix can be generated using the information, and a fifth matrix corresponding to the fruit and vegetable-gene activity information can be obtained through matrix operations using the first to fourth matrices.

For example, referring to Figure 3, the first matrix has each of the fruits and vegetables as a row, the nutrients contained by each fruit and vegetables as a column, and the content of the nutrients contained by each of the fruits and vegetables. It may be a matrix with component values. At this time, it may be desirable to standardize the component values constituting the first matrix using the average and variance values of all component values.

The second matrix may be a matrix that has each nutrient as a row, metabolites metabolized by each nutrient as columns, and the amount of each nutrient metabolized and absorbed as a component value. At this time, it may be desirable to standardize the component values constituting the second matrix using the average and variance values of all component values.

The third matrix is a matrix that has each metabolite as a row, genes that are activated or suppressed when each metabolite is absorbed into the body as columns, and the component value is the degree of activation or suppression by each metabolite. It can be. For example, in the third matrix, the component values of the column corresponding to each of the genes activated by a specific metabolite are assigned +1, and the component values of the column corresponding to each of the genes repressed by a specific metabolite are assigned to -1. You can have it.

The fourth matrix may be a diagonal matrix with genes arranged in the same order in each row and column, and diagonal component values with weight values corresponding to the genes for each individual. In other words, if the user has, or is presumed to have, a gene placed in any i-th row and i-th column, the weight value corresponding to the i-th row and i-th column may be a value greater than 0, and the user according to the lifelog The size of the weight value can be determined by reflecting the activity record. For example, if it is known that a specific gene acts to increase the expression of diabetes, and it is advantageous to walk a lot to reduce the incidence of diabetes, the higher the number of steps according to the life log, the higher the size of the weight value corresponding to the gene. can be set.

Additionally, if the user does not have, or is presumed not to have, a gene placed in any i-th row and i-th column, the weight value corresponding to the i-th row and i-th column may be 0.

At this time, it may be desirable to structure the component values constituting the fourth matrix so that they are not standardized so that the expression or retention of genes according to each individual is best reflected.

The vegetable-gene activity information generating unit 304 may sequentially perform a matrix multiplication operation on the first to fourth matrices, and obtain a fifth matrix corresponding to the fruit vegetable-gene activity information as a result of the matrix multiplication operation.

Referring to FIG. 3, the fifth matrix has, for a specific user, each of the fruits and vegetables as a row, each of the genes as a column, and a value representing the correlation between genes corresponding to each of the fruits and vegetables as component values. there is. That is, a value representing the correlation between genes corresponding to each fruit or vegetable can be determined through a matrix multiplication operation between the first to fourth matrices described above.

The fruit and vegetable-gene activity information generating unit 304 may provide the fruit and vegetable-gene activity information obtained as the fifth matrix to the user terminal 100.

FIG. 4 is a block diagram showing functional components of the user terminal according to FIG. 1.

Referring to FIG. 4, the user terminal 100 collects a lifelog by linking with at least one wearable device 200 or through user input, and provides the collected lifelog to the cloud server 300. Lifelog collection unit 101; an activity information acquisition unit 102 that acquires fruit and vegetable-gene activity information by communicating with the cloud server 300 over a wireless network; an artificial neural network learning unit 103 that supervised learning of the artificial neural network 10 using pre-collected training data; A fruit and vegetable recommendation unit (104) that inputs a number of input vectors obtained from fruit and vegetable-gene activity information into the supervised learning artificial neural network (10) and selects a number of fruits and vegetables to recommend to the user based on the output of the artificial neural network (10). ); and a fruit and vegetable compatibility determination unit 105 that determines compatibility between a plurality of selected fruits and vegetables and provides a subscription service to the user according to the determined compatibility.

Here, the training data uses each of the personalized fruit and vegetable-gene activity information obtained from a large number of different users as input data to the artificial neural network 10, and artificially selects the fruit and vegetable most purchased or viewed by the large number of other users. It can be used as output data for the neural network (10).

The fruit and vegetable recommendation unit 104 sequentially inputs each row constituting the fifth matrix corresponding to the fruit and vegetable-gene activity information as input vectors to the artificial neural network 10, and for each of the input input vectors, Output values of the artificial neural network 10 can be obtained.

Here, each row constituting the fifth matrix corresponds to each fruit and vegetable because it represents the genetic relationship for each fruit and vegetable. Therefore, the output values obtained as the output of the artificial neural network 10 may correspond to the probability of recommending the fruit or vegetable corresponding to each row input as the input vector.

In summary, the fruit and vegetable recommendation unit 104 can select fruits and vegetables corresponding to a preset number of output values in order of largest size among the obtained output values as fruits and vegetables to be recommended to the user.

The fruit and vegetable compatibility determination unit 105 can calculate the degree of compatibility between a plurality of selected fruits and vegetables. For example, the user terminal 100 groups a plurality of selected fruits and vegetables into sets consisting of between 2 and n (n is the total number of selected fruits and vegetables) fruits and vegetables, and groups the fruits and vegetables within each set. The compatibility between the two is calculated for each set, and among the sets where the calculated compatibility is above a preset threshold, the fruits and vegetables in the set selected according to the preset criteria can be provided to the user as fruits and vegetables for the subscription service. .

In order to calculate the compatibility, the fruit and vegetable compatibility determination unit 105 has each fruit and vegetable as a row, each color, seasonal period, harvest amount, and sales price as a column, and an RGB pixel value representing the color and a month corresponding to the seasonal period. A sixth matrix can be created whose component values include the timing value representing (month), harvest amount, and sales price.

At this time, it is desirable that each column constituting the sixth matrix be standardized on a per-column basis using the averages and standard deviations of the components constituting each column.

Next, the fruit and vegetable compatibility determination unit 105 may generate a fruit and vegetable characteristic matrix by adding the columns constituting the sixth matrix to the rear column of the first matrix corresponding to the content information. In other words, in the fruit and vegetable characteristic matrix, each row corresponds to one fruit and vegetable, and each column corresponds to nutrient content, color, season, harvest, and selling price.

The fruit and vegetable compatibility determination unit 105 can calculate the compatibility between fruits and vegetables using the fruit and vegetable characteristic matrix generated in this manner.

For example, when calculating the compatibility between a first fruit and vegetable and a second fruit and vegetable, the fruit and vegetable compatibility determination unit 105 selects a first row vector corresponding to the first fruit and vegetable and a first row vector corresponding to the second fruit and vegetable in the fruit and vegetable characteristic matrix. Compatibility can be calculated by extracting the second row vector and calculating the first row vector and the second row vector according to Equation 1 below.

In the same way, when calculating the compatibility between the first to qth fruits and vegetables (compatibility between q fruits and vegetables), the first to qth row vectors corresponding to the first to qth fruits and vegetables in the fruit and vegetable characteristic matrix The q row vector can be extracted and the compatibility degree calculated according to Equation 1 below.

In Equation 1, Vi and Vj may be the i-th and j-th row vectors selected from the first to q-th row vectors.

The fruit and vegetable compatibility determination unit 105 may select a set according to preset criteria. For example, within the set, the set that has the largest sum of the number of fruits and vegetables that are in season and the number of fruits and vegetables whose sales price is within the average purchase price and a preset error range can be selected.

FIG. 5 is a diagram showing the structure of an artificial neural network linked to the user terminal according to FIG. 1.

Referring to FIG. 5, the artificial neural network 10 includes an input layer 11 that sequentially receives each of the input vectors and consists of a number (n) of input nodes equal to the number of component values of each of the input vectors. A hidden layer 12 that transmits the output vector (Y') obtained using the output values received from the input layer 11 to the output layer 13, and an activation function is applied to the output vector (Y') to generate the output vector (Y') ) may include an output layer 13 that outputs a probability (p) corresponding to ).

At this time, the artificial neural network 10 is supervised in advance using pre-collected training data, and as described above, the training data set is the first corresponding to each of the fruit and vegetable-gene activity information obtained for a number of different users. 5 The input vectors obtained from the rows of the matrix are used as training inputs for the artificial neural network (10), and the target output vector (Y) corresponding to the fruits and vegetables most purchased or viewed by the majority of other users is used as the training input for the artificial neural network (10). It may be a set of data used as training output for . That is, the target output vector (Y) may be a vector that is uniquely set for each fruit and vegetable.

At this time, when the target output vector (Y) is set in a simple 1:1 matching method for each fruit and vegetable without distinction, even fruits and vegetables with very different types or contents have a similar target output vector (Y) form. This results in very large prediction errors when training data is insufficient.

To solve this problem, the artificial neural network learning unit 103 may refer to content information and previously set a target output vector (Y) corresponding to each fruit or vegetable. For example, the artificial neural network learning unit 103 can generate a first diagonal matrix (F) that satisfies Equation 2 below by diagonalizing the first matrix (K) corresponding to the content information.

In Equation 2, Q is a matrix that reversibly satisfies the relationship according to Equation 2 between the first diagonal matrix (F) and the first matrix (K), and corresponds to a coordinate transformation matrix. The operation according to Equation 2 may be referred to as one of the diagonalization operations, and since it can be easily understood by those skilled in the art, detailed description will be omitted.

Next, the artificial neural network learning unit 103 selects the maximum and minimum values among the diagonal component values constituting the diagonal component of the first diagonal matrix F, and sets an interval between the selected maximum and minimum values. can be equally divided into a number of sections equal to the number of rows of the diagonal matrix (F).

Next, the artificial neural network learning unit 103 converts the first diagonal matrix F into the second diagonal matrix by replacing each of the diagonal components with the minimum value of the section to which each of the diagonal components belongs among the multiple equally divided sections. It can be converted into a matrix, and the row vector corresponding to each of the rows constituting the converted second diagonal matrix can be set as the target output vector corresponding to each of the fruits and vegetables.

When setting the target output vector in the above-described manner, the target output vector for each fruit or vegetable is set differently depending on the content, and at the same time, the component values of the target output vector are determined at as evenly spaced as possible, so the prediction accuracy of the artificial neural network (10) can be quickly improved. It has the advantage of improving.

Meanwhile, if we explain the operation of the artificial neural network 10 again, when the artificial neural network 10 sequentially receives input vectors (D1 to DN) provided as training input values, the output obtained as the output of the hidden layer 12 A loss function is calculated using the vector (Y') and the target output vector (Y) provided as the training output value, and supervised learning is performed to minimize the result of the calculated loss function.

For example, the loss function (H(Y,Y')) may be a cross entropy function. The cross entropy (H(Y,Y')) between the output vector (Y') and the target output vector (Y) can be defined as Equation 3 below.

In Equation 3, Ym may be the mth component (m is a natural number greater than or equal to 1) of the target output vector (Y), and Y'm may be the mth component of the output vector (Y').

Meanwhile, the cross entropy function is widely used as a loss function in a typical artificial neural network (10), but when the target output vector (Y) is configured to correspond to each fruit and vegetable as in the present invention, the correspondence relationship is reflected in the loss function. If this is not done, problems may arise that reduce learning effectiveness.

In order to compensate for this problem, in one embodiment of the present invention, the loss function can be newly defined as in Equation 4 below.

In Equation 4, Y`` is a target output vector predefined to correspond to each of the remaining fruits and vegetables in addition to the fruit and vegetable corresponding to the target output vector (Y), Y1 to Yj (j is 1 smaller than the total number of fruits and vegetables) value). In addition, N may mean the number of component values of the target output vector, Ym is the m (m is a natural number greater than or equal to 1) component of the target output vector (Y), and Y'm is the m component of the output vector (Y'). It may be the second ingredient. Additionally, in Equation 3, UL(Y, Y``) is the Euclidean distance ( Euclidian distance).

That is, because the loss function calculates and reflects the Euclidean distance between the target output vectors corresponding to each fruit and vegetable, the correspondence relationship between the target output vectors corresponding to each fruit and vegetable is taken into consideration in the loss function. There is an effect of additionally improving output accuracy due to learning.

The input layer 11 sequentially receives input vectors D1 to Dn, and applies one or more connection strength values corresponding to the input nodes to the component values of the input vectors to the hidden layer 12. It can be delivered.

For example, one or more connection strength values corresponding to each of the input nodes can be expressed as a first connection strength matrix (W _NХn ) with a size of NХn. At this time, N may be the same number as the input nodes, and n is set equal to n, which is the number of components of the input vectors (D1 to Dn). The first connection strength matrix (W _NХn ) can be set to a random initial value and then continuously updated through supervised learning.

In summary, _the input layer 11 can transmit the intermediate operation vector ( there is.

The hidden layer 12 generates an output vector (Y′) by applying one or more connection strengths corresponding to each of the hidden nodes to the intermediate operation vector (X) received from the input layer 11, and the generated output vector ( Y`) can be transmitted to the output layer 13.

At this time, one or more connection strength values corresponding to each of the hidden nodes can be expressed as a second connection strength matrix (U _nХN ) with a size of nХN. That is, the second connection strength matrix (U _nХN ) restores the intermediate operation vector (X) mapped into n dimensions back to N dimensions.

Meanwhile, the initial value of the second connection strength matrix (U _nХN ) may be set to an arbitrary value, and then continuously updated so that the output vector (Y') becomes the target output vector (Y), which is a training output value. In other words, the second connection strength matrix (U _nХN ) can be updated as the learning data is continuously supervised.

The output layer 13 can output a probability (p) corresponding to the output vector (Y') by applying an activation function to the output vector (Y') received from the hidden layer (12). The activation function has the effect of converting values with various ranges into probabilities by expanding or reducing them to values between 0 and 1. For example, the activation function may be a LeRU function or a Softmax function, but is not limited thereto.

For example, the artificial neural network 10 may be RNN, CNN, Fast RNN, AutoEncoder, etc., but is not limited thereto.

Figure 6 is a first diagram showing an application operating on a user terminal by an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment. Figure 7 is a second diagram showing an application operating on a user terminal by an artificial intelligence-based personalized fruit and vegetable recommendation system according to an embodiment.

Referring to Figures 6 and 7, the above-described user terminal 100 may be run with an application distributed online in advance through the cloud server 300 or online through another external public server installed in advance, The above-described operations or functions may be interpreted as the operations of these applications.

Specifically, referring to FIG. 7, the user terminal 100 can display a plurality of selected fruits and vegetables to the user in order of ranking according to probability values. At this time, not only the fruits and vegetables for the user, but also each of the family members or friends registered in advance by the user Fruits and vegetables can also be selected and displayed to the user. To this end, the user terminal 100 can obtain the life logs of family members or friends by linking with each user terminal of the family members or friends and provide them to the cloud server 300.

In addition, the intake or type of fruit and vegetable intake can be set or modified for each individual or for family or friends.

In addition, when selecting one of the selected fruits and vegetables displayed on the user terminal 100, not only information about the producer, production area, efficacy, and nutrients (nutritional components) of the selected fruit or vegetable is provided, but also useful life information related to the handling or selection of the fruit or vegetable. Information or expert-curated information can be displayed to users.

In addition, icons indicating whether the selected fruit or vegetable are organic, pesticide-free, HACCP certified, suitable for consumption by pets, or suitable for consumption by babies can be displayed to help users easily identify information about the consumption of the fruit or vegetable.

FIG. 8 is a diagram illustrating an exemplary hardware configuration of the user terminal according to FIG. 1.

Referring to Figure 8, the user terminal 100 includes at least one processor 110; And it may include a memory 120 that stores instructions that instruct the at least one processor 110 to perform at least one operation.

Here, at least one operation is interpreted as including at least some of the operations of the user terminal 100 or the functions of the functional units described above.

Here, the at least one processor 110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. You can.

The memory 120 may be comprised of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory 120 may be comprised of at least one of read only memory (ROM) and random access memory (RAM).

The user terminal 100 may include a storage device 160 that stores intermediate processing data, result data, temporary data, etc. obtained while performing the above-described operations. For example, the storage device 160 may be a hard disk drive (HDD), a solid state drive (SSD), or the like.

Additionally, the user terminal 100 may include a transceiver 130 that performs communication through a wireless network. Additionally, the user terminal 100 may further include an input interface device 140, an output interface device 150, etc. Each component included in the user terminal 100 is connected by a bus 170 and can communicate with each other.

Methods according to the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. Computer-readable media may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on a computer-readable medium may be specially designed and constructed for the present invention or may be known and usable by those skilled in the computer software art.

Examples of computer-readable media may include hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions may include machine language code such as that created by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The above-described hardware device may be configured to operate with at least one software module to perform the operations of the present invention, and vice versa.

Additionally, the above-described method or device may be implemented by combining all or part of its components or functions, or may be implemented separately.

Although the present invention has been described above with reference to preferred embodiments, those skilled in the art may make various modifications and changes to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below. You will understand that you can do it.

10: Artificial neural network
100: user terminal
200: Wearable device
300: Cloud server
310: Fruit and vegetable DB
320: Nutrient DB
400: external server
1000: Personalized fruit and vegetable recommendation system based on artificial intelligence

Claims (6)

As an artificial intelligence-based personalized fruit and vegetable recommendation system,
A user terminal that collects the user's lifelog, provides it to a cloud server, and selects personalized fruits and vegetables based on personalized fruit and vegetable-gene activity information obtained from the cloud server; and
It includes a cloud server that generates personalized fruit and vegetable-gene activity information by linking with a pre-built fruit and vegetable database, a nutrient database, and an external server,
The user terminal is,
A plurality of input vectors are generated using the fruit and vegetable-gene activity information, the generated plurality of input vectors are sequentially input into a pre-supervised learning artificial neural network, and the user is generated based on the output value of the artificial neural network. Select a number of fruits and vegetables to recommend to
The fruit and vegetable database stores content information indicating the content of various nutrients contained in each fruit and vegetable,
The nutrient database includes metabolic amount information indicating the amount of nutrients each nutrient is metabolized by the body's metabolic process and absorbed into the body; and
Stores activity information indicating the degree to which each metabolite suppresses or activates gene mutation or gene expression when absorbed into the body;
The cloud server is,
Obtain medical information by linking with the external server, determine weight values for each gene corresponding to the user based on the medical information and the life log, and generate gene weight information representing the determined weight values,
The cloud server is,
A first matrix is generated using the content information, a second matrix is generated using the metabolic amount information, a third matrix is generated using the activity information, and a fourth matrix is created using the gene weight information. Generating and obtaining a fifth matrix corresponding to the fruit and vegetable-gene activity information through a matrix multiplication operation between the first matrix to the fourth matrix,
The user terminal is,
The selected plurality of fruits and vegetables are grouped into sets of fruits and vegetables ranging from 2 to n (n is the total number of the selected plurality of fruits and vegetables), and the compatibility between the fruits and vegetables belonging to each of the sets is calculated. An artificial intelligence-based individual that calculates for each set and provides fruits and vegetables in the set selected according to preset criteria among the sets where the calculated compatibility is more than a preset threshold to the user as fruits and vegetables for a subscription service. Customized fruit and vegetable recommendation system. delete delete delete In claim 1,
The artificial neural network is supervised in advance using training data collected in advance,
The training data uses input vectors obtained from the rows of the fifth matrix corresponding to each of the fruit-gene activity information obtained for a number of users other than the user as training input values for the artificial neural network, and An artificial intelligence-based personalized fruit and vegetable recommendation system that is a set of data that uses the target output vector corresponding to the fruit and vegetable most purchased or viewed by multiple other users as a training output for the artificial neural network. delete