KR20210028554A - Method and apparatus for training AI model using user's time series behavior data - Google Patents

Abstract

According to one disclosure of the present invention, a method for training an AI model using time series behavior data of a user may be provided. The method comprises: a step of acquiring behavioral data for all time-series behaviors of users, who have accessed an integrated platform; a labeling step of setting a label by determining whether target data determined by a manager of the integrated platform has been reached for each of all behavior data of the user; and learning the AI model by repeating the labeling step for all time-series behavioral data of other users accessing the integrated platform.

Description

TECHNICAL FIELD [Method and apparatus for training AI model using user's time series behavior data]

Disclosed is a method and apparatus for training an artificial intelligence model for determining a probability of reaching target data of an integrated platform using all behavioral data of a user who has accessed the integrated platform by one disclosure.

Artificial intelligence (AI) is a branch of computer science and information technology that studies how computers can do the thinking, learning, and self-development that human intelligence can do. It means being able to imitate behavior.

Due to the widespread expansion of new types of multimedia contents, social network service (SNS), and the spread and use of smart devices, the scale of data generated and distributed on the web is increasing exponentially. The vast amount of data that exists and continues to grow on the web can be used to interpret the world. This is big data, which simply means a vast amount of digitized information. By filtering out unnecessary data from big data and extracting and analyzing only useful information, it is possible to read people's thoughts, opinions, and trends, and further predict their behavior in advance. Big data is one of the next generation IT (information technology) technologies that are currently in the limelight not only in Korea but also around the world due to such usefulness.

The domestic big data market is expected to reach KRW 300 billion in 2015 and is expected to grow to KRW 1 trillion in 2020. The size of the domestic market related to big data is also growing by more than 28% every year. However, although big data analysis is explosively increasing, the development speed of data preprocessing technology is as slow as the development of big data analysis technology, and accordingly, the need for the development of automated data preprocessing technology is emerging.

In addition, as the purpose of providing and analysis elements are diversifying according to diversified online services, there is a limitation in analyzing and processing the data desired by users only by the method of analyzing big data. In the case of analysis, there was a limitation in that it was impossible to perform analysis of various dimensions.

Accordingly, the need for an artificial intelligence model capable of performing multidimensional data analysis through informal data collection has increased.

0001) Korean Patent Publication No. 10-2016-0075971 (Publication date: 2016. 06. 30) 0002) Korean Patent Publication No. 10-2019-0062848 (Publication date: 2019.06.07)

The technical task according to the start of work is to collect all the user's behavior data performed from the point of access to the integrated platform to the point of termination, and analyze the user's behavior pattern by labeling whether the collected behavior data corresponds to the purpose data. It is to provide a method and apparatus for training an artificial intelligence model.

Acquiring behavioral data on all time-series behaviors of a user who has accessed the integrated platform according to the first embodiment, and for each of all behavioral data of the user, it is determined whether or not the target data determined by the administrator of the integrated platform has been reached. User time-series behavior data including the step of learning an artificial intelligence model by repeating the labeling step of setting a label and the labeling step for all time-series behavior data of other users who have accessed the integrated platform. It can provide a method for training an artificial intelligence model using.

According to the second embodiment, the processor includes a processor and a memory storing instructions executable by the processor, and the processor executes the instructions, thereby acquiring all time-series behavior data of a user accessing the integrated platform, and each of all behavior data of the user. For the artificial platform, it is determined whether or not the target data determined by the administrator of the integrated platform has been reached, a label is set, and the label is repeatedly set for all time-series behavior data of other users who have accessed the integrated platform. It is possible to provide an apparatus for learning an artificial intelligence model using time-series behavior data of a user, which is to train an intelligence model.

A non-transitory computer-readable storage medium in which instructions executable by a processor are stored in order to execute a method of learning an artificial intelligence model using time-series behavior data of a user according to the third embodiment, and a computer recorded in the storage medium You can provide a program.

By tracking the user's behavior at the start of work, collecting and analyzing user-centered data, it is possible to derive data analysis results of various dimensions that optimize the user's experience.

By automating the data analysis results, which are influenced by the capabilities of the existing practitioners by the start of work, through an artificial intelligence learning model, decision-making in a fast cycle and analysis results of consistent standards according to the automatic data analysis process consisting of data analysis, execution, and strategy. Can provide. In particular, due to the present invention, the working hours of practitioners can be reduced by about 95%, and the number of personnel required for data analysis can also be reduced by 2.8, providing excellent economic effects.

Unlike the existing analysis using non-identifying data by one disclosure, by collecting the profile of an actual user and analyzing all user behaviors, more direct personalized communication is possible and an intimate brand experience can be provided.

According to one disclosure, compared to a relational database that stores data in the form of a table, a graph database that stores each data as an object is advantageous for integrated storage and management of structured and unstructured data.

According to the data analysis system and method of the present invention according to the disclosure, a completed graph data model can be automatically generated by automatically finding a correlation between data when collecting and pre-processing structured and unstructured data, and automatically generating a correlation relationship. Through this, it can be used for product recommendation, user behavior prediction, and intrusion accident prevention analysis advancement by achieving systemic insight that could not be seen in the existing table-type two-dimensional data.

1 is a diagram schematically illustrating a method of training an artificial intelligence model using time-series behavior data of a user according to an exemplary embodiment.
2 is a flowchart illustrating a method of training an artificial intelligence model using time-series behavior data of a user according to an initiation.
3 is a diagram for explaining a feature of learning an artificial intelligence model using graph data according to one disclosure.
FIG. 4 is a diagram for describing a feature of learning an artificial intelligence model using behavioral data to which a success label is assigned according to an initiation.
5 is a diagram for explaining a feature of learning an artificial intelligence model using behavior data to which a failure label is assigned due to an initiation.
6 is a diagram for explaining a feature of acquiring a user behavior according to an initiation and analyzing behavior data by mapping it with user information.
FIG. 7 is a diagram for explaining a feature of managing a user group according to a user's use of an integrated platform according to an initiation.
FIG. 8 is a diagram for explaining the structure of an apparatus for training an artificial intelligence model using time-series behavior data of a user.

The terms used in the present specification will be briefly described, and an embodiment of the present invention will be described in detail.

Terms used in the present specification have selected general terms that are currently widely used as possible while taking functions of the present invention into consideration, but this may vary according to the intention or precedent of a technician working in the field, the emergence of new technologies, and the like. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meaning of the terms will be described in detail in the description of the corresponding invention. Therefore, the terms used in the present specification should be defined based on the meaning of the term and the overall contents of the present invention, not a simple name of the term.

When a part of the specification is said to "include" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. In addition, terms such as "... unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software. .

Machine learning is a branch of artificial intelligence, a field of research that gives computers the ability to learn without explicit programming.

Specifically, machine learning can be said to be a technology that studies and builds a system that learns based on empirical data, performs prediction, and improves its own performance, and algorithms for it. Rather than executing strictly defined static program instructions, machine learning algorithms build specific models to derive predictions or decisions based on input data.

The artificial neural network can be trained using training data. Here, learning means a process of determining parameters of an artificial neural network using training data in order to achieve the purpose of classifying, regressing, or clustering input data. I can. Representative examples of parameters of an artificial neural network include weights applied to synapses or biases applied to neurons.

The artificial neural network learned by the training data may classify or cluster input data according to a pattern of the input data.

Learning methods of artificial neural networks can be broadly classified into supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

Supervised learning is a method of machine learning to infer a function from training data. And among the functions to be inferred, outputting a continuous value is called regression, and predicting and outputting a class of an input vector can be called classification.

In supervised learning, an artificial neural network is trained with a label for training data.

Here, the label may mean a correct answer (or result value) that the artificial neural network must infer when training data is input to the artificial neural network.

In addition, in this specification, setting a label on training data for learning an artificial neural network is referred to as labeling the training data with labeling data.

In this case, the training data and the label corresponding to the training data) constitute one training set, and may be input to the artificial neural network in the form of a training set.

Meanwhile, the training data represents a plurality of features, and labeling of the training data may mean that a label is attached to the feature represented by the training data. In this case, the training data may represent the characteristics of the input entity in the form of a vector.

The artificial neural network can infer a function for the correlation between the training data and the labeling data by using the training data and the labeling data. In addition, parameters of the artificial neural network may be determined (optimized) through evaluation of a function inferred from the artificial neural network.

Unsupervised learning is a type of machine learning, where no labels are given for training data.

Specifically, the unsupervised learning may be a learning method of training an artificial neural network to find and classify patterns in the training data itself, rather than an association relationship between training data and a label corresponding to the training data. The integrated platform may be a web platform or a mobile platform including applications, programs, web pages, etc. installed or driven in the user terminal.

Here, the web browser is a program that enables you to use the Web (WWW: World Wide Web) service, which means a program that receives and displays hypertext described in HTML (hyper text mark-up language). For example, Netscape , Explorer, chrome, etc. In addition, an application means an application program on a terminal, and includes, for example, an app running on a mobile terminal (smart phone). Apps can be downloaded and installed from the application market, a virtual marketplace for freely buying and selling mobile content.

The apparatus 100 for training an artificial intelligence model using time-series behavior data of a user according to one disclosure will be abbreviated as the artificial intelligence model learning apparatus 100 and described.

1 is a diagram schematically illustrating a method of training an artificial intelligence model using time-series behavior data of a user according to an exemplary embodiment.

By the start, the artificial intelligence model learning apparatus 100 may collect all behavioral data of a user who has accessed an integrated platform, for example, a shopping site. Here, the behavioral data is visited traffic data connected to the integrated platform, reaction data for content included in the integrated platform, reaction data for events provided by the integrated platform, identification data of users registered on the integrated platform, and access to the integrated platform. It may contain unstructured data that occurs during.

In more detail, the artificial intelligence model learning device 100 includes the path of the user who accessed the website, the connection start time point, the connection end time point, the scroll range of the page, whether or not to log in, the user's content purchase type, whether the purchase step is in progress, and delivery. All formal and unstructured data such as information and payment information can be obtained.

As shown in FIG. 1, the artificial intelligence model learning apparatus 100 may acquire all behavioral data until purchase completion of user A accessed through an SNS advertisement. The artificial intelligence model learning apparatus 100 may acquire user behavior data in real time.

According to an initiation, the artificial intelligence model learning apparatus 100 may automatically determine whether the user's behavior data includes target data previously designated by the administrator of the integrated platform. For example, when the target data of the shopping mall site is “purchase”, the artificial intelligence model learning apparatus 100 may automatically determine whether purchase data exists in all behavioral data of user A. The artificial intelligence model learning apparatus 100 may assign a success label to all behavior data of user A when purchase data exists in the behavior data of user A. That is, the artificial intelligence model learning apparatus 100 may determine all the data acquired from the time point A accesses the website to the time point where it ends as the success label.

By an initiation, the artificial intelligence model learning apparatus 100 may generate an artificial intelligence model by iteratively learning behavioral data that has been subjected to the labeling step. Here, the artificial intelligence model can generate a list belonging to multidimensional results in that it automatically derives the success or failure from the user's behavior data once the target data is determined, unlike the existing learning only for a specific domain. This has the advantage of analyzing user behavior data and utilizing it as a variety of marketing resources.

In the disclosed embodiment, the apparatus 100 for learning an artificial intelligence model may acquire training data by collecting behavioral data that has been subjected to a labeling step. The artificial intelligence model learning apparatus 100 may train an artificial intelligence model by using the acquired training data.

For example, the artificial intelligence model learning apparatus 100 may perform training at a preset period or whenever a preset amount of training data is collected, and iteratively learns the behavioral data that passed the labeling step as described above. It may be updated.

In one embodiment, the learned artificial intelligence model and the module for updating the artificial intelligence model are generated in the main server, and then provided to the artificial intelligence model learning device 100 in various forms (for example, API and SDK, etc.) Can be. The module initially provided may be in a state where basic learning has been completed, and in addition to this, it will be configured in the form of acquiring and learning learning data based on the goals entered by the administrator of the integrated platform, previously collected information, and information collected later. I can.

A customized event based on the user's behavior data can be created using the artificial intelligence model generated by the initiation of the day, and in more detail, a customized event is provided to the user based on the acquisition time of the user's behavior data and the path of the behavior data. It provides the effect of inducing the user's actions to target data.

In addition, it is possible to determine a potential customer to provide a customized event by using the artificial intelligence model generated by the initiation of the day. For example, if, as a result of analyzing the user's behavior data using an artificial intelligence model created at the beginning of the day, that the probability that the user will perform an action corresponding to a preset purpose is greater than or equal to the preset probability, the user Can be judged as a potential customer.

2 is a flowchart illustrating a method of training an artificial intelligence model using time-series behavior data of a user according to an initiation.

In block 201, the apparatus 100 for learning an artificial intelligence model according to an initiation may acquire behavior data for all time-series behaviors of a user accessing the integrated platform.

The artificial intelligence model learning apparatus 100 according to one initiation may collect user's behavior data separately from the user's identification information, map and store the user's identification information, or separately store only the user's behavior data. .

The apparatus 100 for learning an artificial intelligence model according to an initiation may simultaneously acquire time, location, access environment information, network usage information, etc. of each occurrence of the user's behavior data.

In block 202, the apparatus 100 for learning an artificial intelligence model according to one start is a labeling step of setting a label by determining whether the target data determined by the administrator of the integrated platform has been reached for each of the user's behavioral data. You can do it.

Labeling may be determined as success or failure based on whether all behavioral data of the user has reached the target data. In other words, all of the user's behavior data is regarded as failure, except for the case in which purpose data to be assigned a success label exists.

In some cases, labeling may have various result values other than success/failure. For example, in order to derive various marketing methods of a shopping mall site, preset results such as product purchase, membership registration, follow, like, comment, and email subscription may be labeled.

Here, the label may mean a correct answer (or result value) that the artificial neural network must infer when training data is input to the artificial neural network. In addition, for training of an artificial neural network, labeling of training data is referred to as labeling of training data with labeling data.

In this case, the training data and the label corresponding to the training data constitute one training set, and may be input to the artificial neural network in the form of a training set. Meanwhile, the training data represents a plurality of features, and labeling of the training data may mean that a label is attached to the feature represented by the training data. In this case, the training data may represent the characteristics of the input entity in the form of a vector.

The artificial neural network can infer a function for the correlation between the training data and the labeling data by using the training data and the labeling data. In addition, parameters of the artificial neural network may be determined (optimized) through evaluation of a function inferred from the artificial neural network.

In block 203, the apparatus 100 for learning an artificial intelligence model according to an initiation may train the artificial intelligence model by repeating the labeling step for all time-series behavior data of other users accessing the integrated platform.

According to an initiation, the artificial intelligence model learning apparatus 100 may acquire target data from the administrator of the integrated platform. The target data is at least one behavioral data designated by the administrator of the integrated platform, and the artificial intelligence model is individually learned based on the attributes of the target data. For example, the purpose data of the shopping mall site may be "purchase of contents". As another example, the purpose data of the shopping mall site may be variously set, such as membership registration, follow, likes, comments, and email subscriptions according to settings, but is not limited thereto.

As another example, the purpose data of the shopping mall site may include a user's revisiting and repurchasing. In this case, after collecting the user's behavioral data, data on whether the user revisits the site and whether to repurchase accordingly may be obtained, and labeling may be performed based on this.

In addition, the artificial intelligence model learning apparatus 100 may set a success label for each of all the user's behavior data when the target data of the integrated platform exists among all the user's behavior data.

The apparatus 100 for learning an artificial intelligence model may set a failure label for each of all the user's behavior data when the target data of the integrated platform does not exist among all the user's behavior data. The artificial intelligence model learning apparatus 100 may train an artificial intelligence model by using labeled behavioral data. In an embodiment, the apparatus 100 for learning an artificial intelligence model may train an artificial intelligence model using behavioral data to which the same label is assigned, but is not limited thereto.

3 is a diagram for explaining a feature of learning an artificial intelligence model using graph data according to one disclosure.

According to one disclosure, the apparatus 100 for learning an artificial intelligence model may generate graph data by connecting vertices representing each of the behavioral data to which the same label is set with an edge, which is a vector having a direction.

Graph data is a data storage and management technology, and is used as a means of storing each data object and its relationship. Artificial intelligence learning using graph data stores data as one object compared to a relational database that stores each object in accordance with a predefined schema structure, so it is an integrated storage and management of structured and unstructured data. Is advantageous to

The artificial intelligence model learning apparatus 100 may calculate a probability of occurrence of an edge that may be generated from graph data based on the number of edges connected to the vertices. Here, the edge is a virtual association that connects each vertex, and is an index that checks whether other behavioral data can be predicted from the behavioral data. According to one initiation, calculating the probability of occurrence of an edge may be performed based on a Naive Bayes statistical probability technique.

The artificial intelligence model learning apparatus 100 may derive a correlation between the vertices by generating a new edge in graph data when the probability of occurrence of an edge is greater than or equal to a threshold value. Here, the threshold value is a criterion for determining whether to reflect the correlation in graph data according to the probability of occurrence of the correlation (edge), and may be designated by the administrator. As a result, it is possible to automatically create an association relationship between behavioral data that does not have an explicit association relationship.

In addition, the artificial intelligence model learning apparatus 100 may learn a probability of reaching the target data from the user's behavior data by using the association relationship. In order to automatically generate and learn the relationship by initiation, the vertices corresponding to the data objects that can be checked for the occurrence of the relationship on the graph data, edge information connecting the vertices, and the previously acquired data objects Vertex and edge information with properties similar to those of the fields are required. According to the start, the apparatus 100 for learning an artificial intelligence model may continue learning by merging the behavioral data acquired in real time with the existing data.

Conventionally, a relationship between data objects was derived through the probability of the existence of the frequency number of common topics between different data and specific data such as documents or categories. However, such a conventional method has a limitation in that there is frequently no association between data collected and purified in the process of integrating structured and unstructured data and converting it into graph data.

3 is a visual representation of graph data, where each circle represents a vertex, and an arrow connecting the circles represents an edge, which is a vector corresponding to an association relationship between the vertices. According to an embodiment of the present invention shown in FIG. 3, in order to determine the existence condition of the edge 320 that does not currently exist physically, the probability of the entire graph model G with directionality is reviewed, and this probability and the integrated platform By comparing the threshold value set by the administrator of the graph data, the new relationship data, called the edge 320 between the behavior data 311 and 312, can be automatically derived by integrating the relationship information in the graph data and the probability statistics technique. .

The artificial intelligence model learning apparatus 100 may automatically generate a completed graph data model by automatically finding a relationship between data during collection and preprocessing of structured and unstructured data through artificial intelligence learning based on graph data. In addition, it can be usefully used for product recommendation, predicting user behavior, creating customized events, and deriving marketing methods by achieving systemic insights that could not be seen in existing table-type two-dimensional data through automatic association generation. Furthermore, since it is possible to automatically create associations between data and derive various association patterns within behavioral data, meaningful interpretations using only limited data are possible, which can be usefully used for artificial intelligence preliminary research techniques.

4 is a diagram for explaining a feature of learning an artificial intelligence model by using behavior data to which a success label is assigned according to an initiation.

The artificial intelligence model learning apparatus 100 according to one disclosure may detect a redundant path of the success label by using the similarity of all graph data of all users set as the success label. For example, looking at the success data of users A, B, and C who purchase content on a shopping site, it can be confirmed that after all of A, B, and C have searched for a product, they have searched for a product review of the corresponding product. In other words, it can be seen that after the product inquiry action, the action that continuously performs the product review inquiry action has a high probability of leading to a purchase. It can be seen that the paths of such product inquiry and product review inquiry were duplicated three times, and the identity of the connection path was detected as two times for access through SNS advertisement, two times for login before purchase, and two times for issuance of discount coupons.

The artificial intelligence model learning apparatus 100 may assign a weight to the overlapping path of the success label based on the accumulated number of overlaps. The artificial intelligence model learning apparatus 100 may assign a higher weight as the number of users performing common behavioral data increases. That is, as the number of behavioral data to be performed increases, the person performing the behavioral data has a higher probability of reaching the target data.

In addition, by assigning weights to repeated data among success data, it is possible to derive a multifaceted marketing method of the corresponding integrated platform. For example, it is possible to recommend related products to users who have accessed through a high-weight access path, provide coupons for products using a list of users who have accessed through a low-weight access path, or change the advertising method of the access path. .

The artificial intelligence model learning apparatus 100 determines the shortest path from predetermined behavior data included in the overlapping path of the success label to the target data based on the weight given to the overlapping path of the success label, and a success probability associated with the shortest path. You can create incremental events.

The shortest path by initiation refers to a path of behavior data that can increase the probability of success most based on the current user's behavior. The shortest path may be set differently according to the type of target data, and may be set differently according to the property of the behavior data. For example, when a user acquires behavioral data that spends a lot of time on a detailed description of a first product and a product review, it is possible to induce a purchase of the first product by providing a coupon for the first product to the user. Alternatively, when the page of the first product and the page of the second product are alternately clicked and behavioral data for dedicating time is obtained, an event for discounting when two products are purchased at the same time may be set as the shortest path.

The artificial intelligence model learning device 100 may determine a success probability increase type event using an artificial intelligence learning model, or determine any one of at least one event provided from the administrator of the integrated platform as a success probability increase type event. I can.

FIG. 5 is a diagram for explaining a feature of learning an artificial intelligence model using behavior data to which a failure label is assigned due to an initiation.

The artificial intelligence model learning apparatus 100 may detect the redundant path by using the similarity of all graph data of all users to which the failure label is assigned. For example, using data on failures due to non-purchase of D and E, D and E search for several products for a short time, and use the similarity of the path to other products to search for products and change pages for a short time. Behavioral data can be detected as redundant paths.

The artificial intelligence model learning apparatus 100 may assign a weight to the overlapping path of the failure label based on the accumulated number of overlaps. The artificial intelligence model learning apparatus 100 analyzes the cause of the repeated failure and assigns a weight to it, so that it can be checked for what cause the user leaves.

The artificial intelligence model learning apparatus 100 may generate a failure probability improvement type event based on a weight assigned to a duplicate path of the failure label. For example, when the page bounce rate is relatively high on the product detail information page, the artificial intelligence model learning apparatus 100 may induce a purchase conversion by providing a coupon to the user using a list of users who have ended.

According to an initiation, the apparatus 100 for learning an artificial intelligence model may provide a failure probability improvement event when a duplicate path of a failure label is detected from the user's behavior data. For example, by providing a coupon for a specific product to a user who has a history of ending on a page of a specific product, it is possible to induce a user's purchase conversion.

The artificial intelligence model learning device 100 may determine a failure probability improvement type event using the artificial intelligence learning model, or determine any one of at least one event provided from the administrator of the integrated platform as a failure probability improvement type event. I can.

In an embodiment, the artificial intelligence model learning apparatus 100 may collect and store information about each user's behavioral data, information about an induction plan for achieving a purpose provided accordingly, and information about whether or not the purpose is achieved accordingly. The artificial intelligence model learning apparatus 100 may train an artificial intelligence model that can derive information on guidance measures that can induce the user's goal attainment based on the user's behavior data based on the collected information. .

According to an embodiment, the apparatus 100 for learning an artificial intelligence model may acquire information on a change in an achievement probability after providing an induction plan compared to an existing achievement probability in addition to whether or not an object has been achieved. The artificial intelligence model learning apparatus 100 may perform valid labeling for induction books whose achievement probability is changed (increased) by more than a preset reference value, and invalid labeling for induction books that are not.

The artificial intelligence model learning apparatus 100 may train an artificial intelligence model that determines an induction book to be provided to a user based on the labeled learning data.

In an embodiment, the apparatus 100 for learning an artificial intelligence model may set a difficulty level of an objective to be achieved. For example, when the difficulty of purchasing a product is 10, the difficulty of adding to the cart is 7, the difficulty of subscribing to an email is 5, the difficulty of following a page is 3, and the difficulty of clicking a like is 2, respectively.

Accordingly, when the product purchase is set for the final purpose, the artificial intelligence model learning apparatus 100 determines whether to put the currently viewed product in the shopping cart when the predicted goal achievement probability based on the user's behavior data is 70%. You can provide the user with a query or pop-up page that asks.

Similarly, when the predicted goal achievement probability based on the user's behavior data is 50%, a message requesting an email subscription to the user may be provided. In other words, in the case of a model trained according to the artificial intelligence model learning apparatus 100 and a system using the same, at the same time, in the case of failure to achieve the goal, the maximum result is obtained while providing guidance to the user to achieve the final purpose. By requesting actions corresponding to this to the user, it is possible to induce a positive action and feedback of the user.

In the case of online marketing, it is common for users to interact in real time through offline marketing and to provide a dichotomous result of whether a predetermined purpose is achieved or not achieved, unlike consulting according to the situation. According to the disclosed embodiment, there is an advantage of providing a model capable of obtaining a maximum secondary performance according to the current state of the user even if the final purpose is not achieved based on the user's behavior data.

In addition, the artificial intelligence model learning apparatus 100 may determine the possibility of achieving a goal according to the user's behavior data based on the model according to the disclosed embodiment, and accordingly determine to provide a discount coupon to the user.

For example, the artificial intelligence model learning apparatus 100 may not provide a separate discount coupon when the probability that the user purchases a product is greater than or equal to a first threshold value. On the other hand, when the probability that the user purchases the product is equal to or greater than the second threshold and equal to or less than the first threshold, a discount coupon may be provided to induce the user's purchase.

In an embodiment, the discount rate of the discount coupon may be determined based on the selling price of the same product on different platforms. For example, the artificial intelligence model learning apparatus 100 acquires price information including the lowest Internet price of the product currently viewed by the user and the average price on other platforms, and the selling price on the page currently viewed by the user Can be compared. At this time, the artificial intelligence model learning apparatus 100 is the lowest Internet price among the Internet lowest price of the product and the selling price on the page currently viewed by the user, which corresponds to the middle of the Internet lowest price or the Internet lowest price A discount rate for providing a discount price of a specific cost may be determined, and a discount coupon may be provided accordingly.

6 is a diagram for explaining a characteristic of acquiring a user behavior according to an initiation and analyzing behavior data by mapping it with user information.

According to one disclosure, in FIG. 6(a), the apparatus 100 for learning an artificial intelligence model shows an embodiment in which the user behavior is actually collected. By initiation, the artificial intelligence model learning apparatus 100 may collect behavioral data from a user's access point to an end point of the integrated platform. For example, behavioral data indicating that a user viewed a product page at 8:3 and purchased a product at 8:04 may be collected.

According to one disclosure, FIG. 6(b) shows information of a user who has accessed the integrated platform. User information according to an initiation may include personal identification information and recent visit information of the user.

At the beginning, the artificial intelligence model learning apparatus 100 may determine a success probability based on the user's behavior data by separately storing user information and behavior data and analyzing the data for each. Further, the artificial intelligence model learning apparatus 100 maps, stores, and analyzes user information and user behavior data, thereby enabling user-customized data analysis. Therefore, customized analysis according to the user's behavior pattern becomes possible.

FIG. 7 is a diagram for describing a feature of managing a user group according to a user's use of an integrated platform according to an initiation.

According to one disclosure, the artificial intelligence model learning apparatus 100 may check the user's use of the integrated platform and create a user group suitable for each characteristic.

According to one disclosure, the apparatus 100 for learning an artificial intelligence model may classify users who have a history of reaching target data into a first group based on the user's information on using the integrated platform. For example, the artificial intelligence model learning apparatus 100 may be classified into a one-time purchaser group, a two-time purchaser group, or the like according to the number of purchases by the user. In addition, the artificial intelligence model learning apparatus 100 may create a user group that does not have a previous purchase history, and more specifically, may create a detailed group based on whether or not a purchase has been made based on which behavior data. .

In addition, the artificial intelligence model learning apparatus 100 determines at least one recommended content to be provided to the first group based on the property of the target data previously reached by the first group, and provides at least one recommended content to the first group. Can be provided to. For example, when the user belongs to a group that has purchased at least two or more times, the artificial intelligence model learning apparatus 100 may determine content to be recommended to the user based on the purchased content history. Alternatively, a specific product that the user clicks at least twice or more may be determined as recommended content, and a coupon for a specific product may be issued.

That is, the artificial intelligence model learning apparatus 100 can be used as marketing data by analyzing and grouping all user actions.

FIG. 8 is a diagram for explaining the structure of an apparatus for training an artificial intelligence model using time-series behavior data of a user.

However, not all of the components shown in FIG. 8 are essential components of the apparatus 100 for learning an artificial intelligence model. The apparatus 100 for learning the artificial intelligence model may be implemented by more components than those shown in FIG. 8, or by fewer components.

The processor 1100 generally controls the overall operation of the artificial intelligence model learning apparatus 100. According to one disclosure, the controller is also referred to as a processor. For example, the processor 1100 includes programs stored in the memory 1200. By executing, the user input unit 1100, the output unit 1200, the sensing unit 1400, the communication unit 1500, the A/V input unit 1600 can be controlled overall. By executing programs stored in 1200, the function of the artificial intelligence model learning apparatus 100 may be performed. The processor 1100 may include at least one processor. The processor 1100 may be provided with its functions and roles. Accordingly, a plurality of processors may be included, or an integrated type of processor may be included. In one embodiment, the processor 1100 executes at least one program stored in the memory 1200 to provide a notification message. It may include at least one processor.

The memory 1200 may store a program for processing and controlling the processor 1100, and may store data input to the behavior analysis apparatus 100 or output from the behavior analysis apparatus 100.

The memory 1200 is a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (for example, SD or XD memory 1200). ), RAM (Random Access Memory) SRAM (Static Random Access Memory), ROM (ROM, Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), Magnetic Memory , A magnetic disk, and an optical disk.

Programs stored in the memory 1200 may be classified into a plurality of modules according to their functions. Here, the plurality of modules are software, not hardware, and refer to modules that operate functionally.

The processor 1100 according to an embodiment may include a data learning unit 1101 and a data recognition unit 1102.

At least one of the data learning unit 1101 and the data recognition unit 1102 may be manufactured in the form of at least one hardware chip and mounted on an electronic device. For example, at least one of the data learning unit 1101 and the data recognition unit 1102 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or an existing general-purpose processor (eg, a CPU Alternatively, it may be manufactured as a part of an application processor) or a graphics dedicated processor (eg, a GPU) and mounted on the aforementioned various electronic devices.

In this case, the data learning unit 1101 and the data recognition unit 1102 may be mounted on one electronic device, or may be mounted on separate electronic devices, respectively. For example, one of the data learning unit 1101 and the data recognition unit 1102 may be included in the electronic device, and the other may be included in the server. In addition, the data learning unit 1101 and the data recognition unit 1102 may provide model information built by the data learning unit 1101 to the data recognition unit 1102 through wired or wireless communication, or the data recognition unit ( Data input to 1102 may be provided to the data learning unit 1101 as additional learning data.

Meanwhile, at least one of the data learning unit 1101 and the data recognition unit 1102 may be implemented as a software module. When at least one of the data learning unit 1101 and the data recognition unit 1102 is implemented as a software module (or a program module including an instruction), the software module is computer-readable and non-transitory. It may be stored in a non-transitory computer readable media. Also, in this case, at least one software module may be provided by an operating system (OS) or may be provided by a predetermined application. Alternatively, some of the at least one software module may be provided by an operating system (OS), and the remaining part may be provided by a predetermined application.

The method according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (9)

Acquiring behavioral data on all time-series behaviors of a user accessing the integrated platform;
A labeling step of setting a label by determining whether target data determined by an administrator of the integrated platform has been reached for each of the user's behavior data; And
Including the step of learning an artificial intelligence model by repeating the labeling step for all time-series behavior data of other users who have accessed the integrated platform,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 1,
The labeling step,
Obtaining target data from an administrator of the integrated platform;
Setting a success label for each of all of the user's behavioral data when object data of the integrated platform exists among all the behavioral data of the user;
Setting a failure label for each of the user's all behavioral data when the object data of the integrated platform does not exist among all the behavioral data of the user; And
Including the step of training an artificial intelligence model by using the label-set behavioral data,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 1,
The behavioral data includes visit traffic data accessing the integrated platform, response data for content included in the integrated platform, response data for events provided by the integrated platform, identification data of a user registered in the integrated platform, and the Characterized in that it contains unstructured data generated while accessing the integrated platform,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 1,
The target data is at least one behavioral data designated by an administrator of the integrated platform, wherein the artificial intelligence model is individually learned based on an attribute of the target data,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 1,
The step of training the artificial intelligence model,
Generating graph data by connecting vertices representing each of the behavioral data for which the same label is set with an edge, which is a vector having a directionality;
Calculating a probability of occurrence of an edge that can be generated from the graph data based on the number of edges connected to the vertices;
When the probability of occurrence of the edge is greater than or equal to a threshold value, generating a new edge in the graph data to derive a relationship between the vertices; And
Including the step of learning the probability of reaching the target data from the user's behavior data using the association,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 5,
Detecting a redundant path of the success label by using the similarity of all graph data of all users set as the success label;
Assigning a weight to the duplicate path of the success label based on the accumulated number of duplicates;
Determining a shortest path from predetermined behavioral data included in the overlapping path of the success label to target data based on a weight assigned to the overlapping path of the success label;
Generating a success probability increasing event associated with the shortest path; And
In case a duplicate path of the success label is detected from the user's behavior data, providing the success probability increase type event,
A method of training an artificial intelligence model using the user's time series behavioral data. The method of claim 5,
Detecting a redundant path by using the similarity of all graph data of all users to which a failure label is assigned;
Assigning a weight to the duplicate path of the failure label based on the accumulated number of duplicates;
Generating a failure probability improvement type event based on a weight assigned to a duplicate path of the failure label; And
In case a duplicate path of the failure label is detected from the user's behavior data, providing the failure probability improvement type event,
A method of training an artificial intelligence model using the user's time series behavioral data. A memory for storing one or more instructions; And
A processor that executes the one or more instructions stored in the memory,
The processor executes the one or more instructions,
An apparatus for performing the method of claim 1. A computer program combined with a computer as hardware and stored in a recording medium readable by a computer to perform the method of claim 1.

Images