WO2023033538A1 - 전자 장치 및 전자 장치의 제어 방법 - Google Patents
전자 장치 및 전자 장치의 제어 방법 Download PDFInfo
- Publication number
- WO2023033538A1 WO2023033538A1 PCT/KR2022/013026 KR2022013026W WO2023033538A1 WO 2023033538 A1 WO2023033538 A1 WO 2023033538A1 KR 2022013026 W KR2022013026 W KR 2022013026W WO 2023033538 A1 WO2023033538 A1 WO 2023033538A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- graphs
- augmented
- individual
- graph
- electronic device
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
- H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
Definitions
- the present disclosure relates to an electronic device and a method for controlling the electronic device, and more particularly, to an electronic device capable of providing information on recommended content using a neural network model and a method for controlling the electronic device.
- the present disclosure is to overcome the limitations of the prior art as described above, and an object of the present disclosure is to learn a neural network model based on effectively augmenting an individual graph representing a user's access history to content, and accordingly It is an object of the present invention to provide an electronic device capable of predicting recommended content that meets a user's intention and a method for controlling the electronic device.
- an electronic device obtains a memory for storing a neural network model and a plurality of individual graphs representing a user's access history for a plurality of contents for each of a plurality of sessions, ,
- An integrated graph in which the plurality of individual graphs are integrated is generated based on the connection relationship of the nodes included in the plurality of individual graphs and the number of repetitions of the connection of the nodes, and based on the integrated graph, the plurality of individual graphs and a processor for acquiring a plurality of augmented graphs by augmenting each of the augmented graphs and learning the neural network model for providing recommended content based on the plurality of augmented graphs.
- the processor may obtain at least one augmented graph among the plurality of augmented graphs by adding at least one node included in the integrated graph to a plurality of nodes included in each of the plurality of individual graphs.
- the processor obtains at least one augmented graph among the plurality of augmented graphs by changing at least one node among a plurality of nodes included in each of the plurality of individual graphs to at least one node included in the integrated graph. can do.
- the processor may train the neural network model based on contrastive loss for the plurality of augmented graphs.
- the neural network model includes an encoder that obtains vectors corresponding to the plurality of individual graphs and vectors corresponding to the plurality of augmented graphs, a prediction module that obtains recommended content information corresponding to each of the plurality of individual graphs, and the and an augmentation module that acquires the plurality of augmented graphs by augmenting each of a plurality of individual graphs, and a contrast loss acquisition module that obtains the contrast loss for the plurality of augmented graphs.
- the encoder maps augmented graphs corresponding to the same individual graph among the plurality of augmented graphs to vectors having a short distance in a vector space, and maps the augmented graphs corresponding to the same individual graph among the plurality of augmented graphs to different individual graphs. It can be learned to map corresponding augmented graphs to vectors of long distances in the vector space.
- a vector space for defining vectors corresponding to the plurality of augmented graphs may be different from a vector space for defining the contrast loss for the plurality of augmented graphs.
- the augmentation module acquires three or more augmented graphs for each of the plurality of individual graphs by augmenting each of the plurality of individual graphs in three or more different methods, and the contrast loss corresponds to the three or more augmented graphs. It can include two or more positive pairs defined based on vectors.
- a method for controlling an electronic device using a neural network model for providing recommended content based on a user's access history to content is provided for a user of a plurality of content.
- Generating an integrated graph acquiring a plurality of augmented graphs by augmenting each of the plurality of individual graphs based on the integrated graph, and using the neural network model for providing recommended content based on the plurality of augmented graphs. It includes a learning step.
- the obtaining of the plurality of augmented graphs may include adding at least one node included in the integrated graph to a plurality of nodes included in each of the plurality of individual graphs, so as to obtain at least one augmented graph among the plurality of augmented graphs. It may include the step of obtaining.
- the acquiring of the plurality of augmented graphs may include changing at least one node among a plurality of nodes included in each of the plurality of individual graphs to at least one node included in the integrated graph, so that among the plurality of augmented graphs It may include obtaining at least one augmented graph.
- the neural network model may be trained based on a contrast loss for the plurality of augmented graphs.
- the neural network model includes an encoder that obtains vectors corresponding to the plurality of individual graphs and vectors corresponding to the plurality of augmented graphs, a prediction module that obtains recommended content information corresponding to each of the plurality of individual graphs, and the and an augmentation module that acquires the plurality of augmented graphs by augmenting each of a plurality of individual graphs, and a contrast loss acquisition module that obtains the contrast loss for the plurality of augmented graphs.
- the encoder maps augmented graphs corresponding to the same individual graph among the plurality of augmented graphs to vectors having a short distance in a vector space, and maps the augmented graphs corresponding to the same individual graph among the plurality of augmented graphs to different individual graphs. It can be learned to map corresponding augmented graphs to vectors of long distances in the vector space.
- a vector space for defining vectors corresponding to the plurality of augmented graphs may be different from a vector space for defining the contrast loss for the plurality of augmented graphs.
- the augmentation module acquires three or more augmented graphs for each of the plurality of individual graphs by augmenting each of the plurality of individual graphs in three or more different methods, and the contrast loss corresponds to the three or more augmented graphs. It can include two or more positive pairs defined based on vectors.
- a non-transitory computer-readable recording medium including a program for executing a control method of an electronic device, based on a user's access history to content
- the control method of the electronic device using a neural network model for providing recommended content includes obtaining a plurality of individual graphs representing a user's access history to a plurality of contents for each of a plurality of sessions, and a node included in the plurality of individual graphs.
- FIG. 1 is a flowchart briefly illustrating the configuration of an electronic device according to an embodiment of the present disclosure
- FIG. 2 is a diagram showing a plurality of modules included in a neural network model according to the present disclosure and input/output data for each of the plurality of modules;
- FIG. 3 is a diagram for explaining in detail an augmentation process of individual graphs according to various embodiments of the present disclosure
- FIG. 4 is a block diagram showing the configuration of an electronic device according to an embodiment of the present disclosure in detail
- FIG. 5 is a diagram showing information on a user's access history in detail according to an embodiment of the present disclosure
- FIG. 6 is a diagram showing an example of a user interface for providing information on recommended content based on a user's access history
- FIG. 7 is a diagram showing information on a user's access history in detail according to an embodiment of the present disclosure.
- FIG. 8 is a diagram showing an example of a user interface for providing information on recommended content based on a user's access history
- FIG. 9 is a flowchart illustrating a control method of an electronic device according to an embodiment of the present disclosure.
- expressions such as “has,” “can have,” “includes,” or “can include” indicate the presence of a corresponding feature (eg, numerical value, function, operation, or component such as a part). , which does not preclude the existence of additional features.
- expressions such as “A or B,” “at least one of A and/and B,” or “one or more of A or/and B” may include all possible combinations of the items listed together.
- a component e.g., a first component
- another component e.g., a second component
- connection to it should be understood that the certain component may be directly connected to the other component or connected through another component (eg, a third component).
- the phrase “device configured to” may mean that the device is “capable of” in conjunction with other devices or components.
- a processor configured (or configured) to perform A, B, and C may include a dedicated processor (eg, embedded processor) to perform the operation, or by executing one or more software programs stored in a memory device.
- a general-purpose processor eg, CPU or application processor
- a 'module' or 'unit' performs at least one function or operation, and may be implemented with hardware or software, or a combination of hardware and software.
- a plurality of 'modules' or a plurality of 'units' may be integrated into at least one module and implemented by at least one processor, except for 'modules' or 'units' that need to be implemented with specific hardware.
- 1 is a flowchart briefly illustrating the configuration of an electronic device 100 according to an embodiment of the present disclosure.
- 2 is a diagram showing a plurality of modules included in the neural network model according to the present disclosure and input/output data for each of the plurality of modules.
- the 'electronic device 100' refers to a device capable of providing recommended content based on a user's access history to content.
- the electronic device 100 may provide recommended content using the learned neural network model.
- the electronic device 100 may be a server providing content, but the type of the electronic device 100 according to the present disclosure is not limited to a specific type.
- the 'neural network model' refers to an artificial intelligence model trained to acquire information about recommended content based on information about a user's access history to content.
- the type of neural network model according to the present disclosure for convenience of explanation, a case in which the neural network model includes a graph neural network (GNN) will be described below.
- GNN graph neural network
- a 'graph neural network' refers to a neural network trained to output output data corresponding thereto when a graph is input as input data. That is, a graph neural network refers to a neural network that can be directly applied to a graph.
- a 'graph' is a data structure that includes nodes (or points) and lines (or edges) connecting the nodes.
- the graph may include data corresponding to each node included in the graph and data corresponding to a line representing a connection relationship between nodes.
- Inputting a graph to the neural network model may be used in the same sense as inputting data corresponding to each node included in the graph and data corresponding to a line representing a connection between nodes to the neural network model.
- 'nodes' included in the graph represent each of contents accessed by a user
- a 'line' connecting the nodes means a sequential connection relationship between contents accessed by a user. For example, if a user sequentially watches movie A, movie B, and movie C in a particular session, the graph includes three nodes corresponding to movie A, movie B, and movie C, and also has three nodes corresponding to movie A. A line connecting the node and the node corresponding to movie B and a line connecting the node corresponding to movie B and the node corresponding to movie C may be included.
- the term 'user's access history to content' is used as a generic term for information indicating that a user has accessed in the past or present with respect to various contents provided through the electronic device 100 .
- the user's access history to the content is the content playback history if the content is video content or audio content, the loading history of the web page if the content is a web page showing product information, or the user's input for a specific product. history may be included.
- a user's 'access' to content is used as a generic term for an event indicating that the user of the electronic device 100 has interacted with content provided through the electronic device 100 .
- the user's access may not only be performed in an active manner, such as when a specific content is provided to the user according to a user input for selecting a specific content from among a plurality of contents, but also may be performed in the form of notification or advertisement to the user. It may also be performed in a passive manner, as in the case of providing
- a user's access to content may be distinguished according to a plurality of sessions.
- a 'session' may mean a time interval during which a series of interactions between the electronic device 100 and the user lasted.
- 'sustained' may mean that when interactions between the electronic device 100 and the user continuously occur, a blank period between the continuously occurring interactions is equal to or less than a preset threshold time. That is, a plurality of sessions may be distinguished according to whether or not the interaction between the electronic device 100 and the user continues, which will be described in more detail below.
- the plurality of sessions may be distinguished based on the time at which the user accesses the content or the application through which the user accesses the content. Specifically, when a threshold time elapses after the user accesses the content, the access before the threshold time elapses and the access after the threshold time elapses may be included in different sessions. Also, the access performed through the first application and the access performed through the second application may be included in different sessions. In addition, a criterion for distinguishing a plurality of sessions may be set in various ways.
- an electronic device 100 may include a memory 110 and a processor 120 .
- the neural network model according to the present disclosure includes a plurality of components such as an encoder, a prediction module, an augmentation module, and a contrast loss acquisition module. modules may be included.
- At least one instruction related to the electronic device 100 may be stored in the memory 110 .
- an operating system (O/S) for driving the electronic device 100 may be stored in the memory 110 .
- various software programs or applications for operating the electronic device 100 may be stored in the memory 110 according to various embodiments of the present disclosure.
- the memory 110 may include a semiconductor memory such as a flash memory or a magnetic storage medium such as a hard disk.
- Various software modules for operating the electronic device 100 may be stored in the memory 110 according to various embodiments of the present disclosure, and the processor 120 executes various software modules stored in the memory 110 to execute the electronic device.
- the operation of (100) can be controlled. That is, the memory 110 is accessed by the processor 120, and data can be read/written/modified/deleted/updated by the processor 120.
- the term memory 110 refers to the memory 110, a ROM (not shown) in the processor 120, a RAM (not shown), or a memory card (not shown) mounted in the electronic device 100 (for example, , micro SD card, memory stick).
- the memory 110 includes data on a neural network model, information on layers constituting a plurality of modules, information on weights and parameters constituting each layer, and Input/output data for each of a plurality of modules may be stored.
- the memory 110 may store information about a user's access history to content, individual graphs, integrated graphs, augmented graphs, information about recommended content, and the like.
- various information required within the scope of achieving the object of the present disclosure may be stored in the memory 110, and the information stored in the memory 110 may be updated as received from an external device or input by a user. .
- the processor 120 controls overall operations of the electronic device 100 . Specifically, the processor 120 is connected to the configuration of the electronic device 100 including the memory 110, and by executing at least one instruction stored in the memory 110 as described above, the electronic device 100 You have full control over the action.
- Processor 120 may be implemented in a variety of ways.
- the processor 120 may include an application specific integrated circuit (ASIC), an embedded processor, a microprocessor, hardware control logic, a hardware finite state machine (FSM), a digital signal processor processor, DSP) may be implemented as at least one.
- ASIC application specific integrated circuit
- FSM hardware finite state machine
- DSP digital signal processor processor
- the term processor 120 may be used to include a central processing unit (CPU), a graphic processing unit (GPU), and a main processing unit (MPU).
- CPU central processing unit
- GPU graphic processing unit
- MPU main processing unit
- the processor 120 may load data for a plurality of modules stored in the memory 110 and implement various embodiments according to the present disclosure through the plurality of modules. That is, the plurality of modules may be implemented as software modules, but at least some of the plurality of modules may be implemented as hardware modules included in the processor 120 .
- the processor 120 may provide recommended content corresponding to the user's access history by using a neural network model. Specifically, the processor 120 may provide recommended content corresponding to the user's access history by using an encoder and a prediction module included in the neural network model.
- the encoder may obtain vectors corresponding to individual graphs, and in particular, the encoder may include a graph neural network.
- an 'individual graph' refers to a graph that individually represents a user's access history for a plurality of contents for each session, and is distinguished from an integrated graph and an augmented graph as will be described later.
- an individual graph is the contents (A, B, C, D, and F) accessed by the user in the first session (F ⁇ A ⁇ C ⁇ D ⁇ B) of the user's access history. ), sequential connection relationships between the contents (connections in the order of F, A, C, D, and B), and information on the number of connections between the contents (once each).
- the encoder may output a vector corresponding to the input individual graph.
- the vector is a digitized feature of the entire input graph, and may be referred to as a representation or feature representation. Accordingly, in FIG. 2, the vector output through the encoder is indicated by the symbol r. Meanwhile, the encoder may obtain a vector corresponding to an augmented graph as well as an individual graph, which will be described later.
- the prediction module may obtain recommended content information corresponding to individual graphs. Specifically, when a vector corresponding to an individual graph is input, the prediction module may output recommended content information corresponding to the input vector.
- the recommended content information refers to information for recommending content that is highly likely to be accessed by a user after content included in an individual graph input to an encoder.
- Recommended content information may be output in the form of a vector as shown in FIG. 2, and the dimension of the vector output through the prediction module may be different from the dimension of the vector output through the encoder.
- a vector output through the prediction module may be referred to as a term such as a target item. In FIG. 2, the vector output through the prediction module is indicated by the symbol t.
- the processor 120 compares the obtained recommended content information with label information to obtain a main loss, and based on the main loss, a supervised learning method is performed.
- a neural network model can be trained.
- label information to be compared with recommended content information is indicated by the symbol V.
- the processor 120 may train the neural network model in a direction in which the main loss decreases by performing a back propagation process based on the obtained main loss. That is, when the neural network model is trained based on the main loss, the prediction module is trained to obtain recommended content information capable of reducing the main loss, and the encoder is trained to obtain a vector capable of reducing the main loss.
- 'Main loss' is distinguished from contrastive loss as described later, and may be obtained by calculating a cosine similarity between recommended content information obtained through a prediction module and pre-stored label information.
- various similarity calculation techniques distance calculation techniques
- Euclidean distance or Manhattan distance can be used, of course.
- the processor 120 may not only learn a neural network model based on the main loss obtained by comparing recommended content information with label information, but also obtain an augmented graph by augmenting an individual graph, and augment the augmented graph.
- a neural network model can be trained using a graph.
- the processor 120 may obtain a plurality of individual graphs representing a user's access history for a plurality of contents for each of a plurality of sessions. Specifically, the processor 120 may collect information about a user's access history to a plurality of contents for a predetermined period of time. In addition, the processor 120 may classify the user's access history for each of a plurality of sessions, and obtain a plurality of individual graphs representing the user's access history for each session.
- the processor 120 may generate an integrated graph in which the plurality of individual graphs are integrated based on the connection relationship and the number of connections of nodes included in the plurality of individual graphs.
- An 'integrated graph' refers to a graph that integrally represents a user's access history to a plurality of contents.
- the processor 120 generates an integrated graph including nodes corresponding to contents accessed by users during a plurality of sessions, lines representing connection relationships between the nodes, and information on the number of times the nodes are connected. can do.
- the integrated graph is the contents accessed by the user in a plurality of sessions included in the user's access history (A, B, C, D, E, and F), and sequential connections between the contents. It may include information about relationships and the number of connections between contents. For example, in the integrated graph of FIG. 2, A ⁇ C indicates that the user accessed content A and then content C, and the number 2 assigned to A ⁇ C indicates that the user accessed content A and then accessed content A in the entire session. Indicates that the number of accesses to content C is twice.
- the processor 120 may obtain a plurality of augmented graphs by augmenting each of a plurality of individual graphs based on the integrated graph. Specifically, the processor 120 may acquire a plurality of augmented graphs through an augmentation module, that is, an augmentation module refers to a module capable of obtaining a plurality of augmented graphs by augmenting each of a plurality of individual graphs.
- 'Augmentation' refers to a technology that increases the amount of data by applying various algorithms to a small amount of data.
- the augmentation technology can be used to solve the problem that the performance of the neural network model may deteriorate or the problem of underfitting or overfitting may occur when the amount of training data is insufficient in learning the neural network model. .
- An augmented graph according to the present disclosure is a graph generated based on a plurality of individual graphs, and at least one of nodes included in the graph, a connection relationship between nodes, and the number of connections may be different from the plurality of individual graphs.
- the neural network model acquires a vector that better represents the characteristics of the input graph and predicts recommended content that meets the user's intention. In this case, the augmented graph is acquired. After the description of the process, it will be described in more detail.
- the processor 120 may obtain at least one augmented graph from among the plurality of augmented graphs by adding at least one node included in the integrated graph to a plurality of nodes included in each of the plurality of individual graphs. there is. This may be briefly referred to as an 'injection' technique.
- the node added to the individual graph may be determined based on the connection relationship and the number of connections of nodes connected to the integrated graph. For example, based on the fact that the connection relationship F ⁇ A exists in the integrated graph, the augmentation module sets the individual graph corresponding to the session A ⁇ C ⁇ D in FIG. 2 to the connection relationship F ⁇ A ⁇ C ⁇ D.
- An augmented graph can be obtained by changing to a graph having
- the processor 120 augments at least one of a plurality of augmented graphs by changing at least one node among a plurality of nodes included in each of a plurality of individual graphs to at least one node included in an integrated graph. graph can be obtained. This can be briefly referred to as the 'change' technique.
- nodes that are changed in the individual graph may be determined based on the connection relationship and the number of connections of nodes connected to the integrated graph. For example, based on the fact that the number of connections for the connection relationship C ⁇ D in the integrated graph is two times, the augmentation module creates an individual graph corresponding to the session B ⁇ C ⁇ E in FIG. It is possible to obtain an augmented graph by changing the connection relationship called into a graph.
- the processor 120 selects only some nodes among a plurality of nodes included in each of a plurality of individual graphs, excludes some nodes, or rearranges the order of some nodes.
- An augmented graph can be obtained.
- the processor 120 may train a neural network model based on the plurality of augmented graphs. Specifically, the processor 120 may acquire a contrast loss through a contrast loss acquisition module, and train a neural network model according to a contrastive learning method based on the obtained contrast loss, that is, obtain a contrast loss.
- the module refers to a module capable of acquiring contrast loss for a plurality of augmented graphs.
- 'Contrastive learning' is a type of self-supervised learning method. After obtaining augmented data by applying different augmentation techniques to data, the distance between feature representations of positive pairs among the data is close. It refers to a method of learning a neural network model so that the feature representation of a negative pair among the data becomes farther away. Also, 'contrast loss' may be defined as a set of positive and negative pairs representing the relationship between augmented data. The specific meaning of the positive pair and the negative pair will be described later.
- the processor 120 may obtain vectors (ie, representations) corresponding to each of the plurality of augmented graphs by inputting the plurality of augmented graphs to an encoder.
- r a1 , r a2 , and r a3 are vectors corresponding to augmented data obtained by the first augmentation method, augmented data obtained by the second augmentation method, and augmented data obtained by the third augmentation method, respectively. Indicates the vector corresponding to .
- the processor 120 configures the contrast loss by inputting the vectors corresponding to each of the plurality of augmented graphs to the contrast loss acquisition module and projecting them into a vector space of another dimension.
- vector can be obtained. That is, a vector space for defining vectors corresponding to a plurality of augmented graphs may be different from a vector space for defining a contrast loss for a plurality of augmented graphs.
- a vector space for defining contrast loss for a plurality of augmented graphs and a vector space corresponding to an output of a prediction module may be different from each other. That is, according to the present disclosure, separate learning having independence in relation to learning based on the output of the prediction module after mapping the vector output through the encoder onto a vector space that is different from the vector space corresponding to the output of the prediction module. By performing, the encoder can be learned more effectively. The learning process based on contrast loss will be described in more detail below.
- a contrast loss representing a relationship to a plurality of augmented graphs can be obtained accordingly.
- z 1 , z 2 , z 3 and z 4 represent the first to fourth sessions included in the user's access history, respectively, and the superscripts a 1 , a 2 and a 3 denote the first augmentation method, respectively;
- the second enhancement method and the third enhancement method are shown. That is, in the example of FIG. 2, z 1 a1 corresponds to the augmented graph obtained by applying the first augmentation method to the individual graph corresponding to the first session, and z 1 a2 applies the second augmentation method to the individual graph corresponding to the first session. It corresponds to the augmented graph obtained by
- the table of FIG. 2 shows a relationship between a plurality of augmented graphs obtained by augmenting a plurality of individual graphs corresponding to the user's access history of FIG. 2 in three ways.
- the relationship between the plurality of augmented graphs is a relationship indicating that they are augmented graphs corresponding to the same individual graph among the plurality of augmented graphs (hereinafter referred to as a positive pair) and an augmented graph corresponding to different individual graphs among the plurality of augmented graphs. It may include a relationship indicating that it is a pair (hereinafter referred to as a negative pair).
- p represents a positive pair and a blank represents a negative pair.
- a relationship between z 1 a1 , z 1 a2 , and z 1 a3 is a positive pair
- a relationship between z 1 a1 , z 2 a1 , z 3 a1 , and z 4 a1 is a negative pair.
- augmented graphs corresponding to the same individual graph (ie, the same session) among a plurality of augmented graphs are set as a positive pair, and different individual graphs (ie, different sessions) among the plurality of augmented graphs are set as positive pairs. ) can be obtained as a negative pair of augmented graphs corresponding to the contrast loss.
- the augmentation module may obtain three or more augmented graphs for each of a plurality of individual graphs by augmenting each of a plurality of individual graphs in three or more different methods.
- contrast loss may include two or more positive pairs defined based on vectors corresponding to three or more augmented graphs.
- the neural network model maps augmented graphs (i.e., positive pairs) corresponding to the same individual graph among a plurality of augmented graphs to vectors having a short distance on the vector space, and maps different individual graphs among the plurality of augmented graphs. It can be learned to map augmented graphs (ie, negative pairs) corresponding to the graph to vectors of long distances on the vector space.
- the neural network model maps z 1 a1 to a vector having a close distance to z 1 a2 and z 1 a3 in a positive pair relationship, and z 2 a1 , z 3 a1 and z 3 a1 in a negative pair relationship. It can be learned to map to a vector that is far from z 4 a1 .
- the process of learning the neural network model based on the recommended content information and the process of learning the neural network model based on the contrast loss may be performed simultaneously or at different times.
- the processor 120 may perform a process of learning a neural network model based on contrast loss and then a process of learning a neural network model based on recommended content information.
- the electronic device 100 can augment an individual graph representing a user's access history to content and effectively learn a neural network model based on the output of the augmented graph.
- the electronic device 100 may acquire a plurality of augmented graphs by using three or more various augmentation methods for individual graphs, and learn an encoder based on a contrast loss including two or more positive pairs. Accordingly, the encoder can obtain a vector that better represents the characteristics of the input graph. In addition, even when an individual graph corresponding to a new session is input to the neural network model during the inference process of the neural network model, it is possible to predict recommended content that meets the user's intention.
- the electronic device 100 may use information about the entire session included in the integrated graph instead of using only information of each session in the process of augmenting an individual graph. Accordingly, in the process of augmenting data It is possible to overcome the problem that only the information inside the session is transferred or part of the information is lost.
- FIG. 3 is a diagram for explaining in detail an augmentation process of individual graphs according to various embodiments of the present disclosure.
- the processor 120 generates an integrated graph in which a plurality of individual graphs are integrated based on the connection relationship and the number of connections of nodes included in the plurality of individual graphs, and each of the plurality of individual graphs is based on the integrated graph.
- a plurality of augmented graphs can be obtained.
- augmenting a plurality of individual graphs based on the integrated graph means obtaining the augmented graph by reflecting information on connection relationships and connection counts of nodes included in the integrated graph to individual graphs.
- FIG. 3 shows exemplary individual graphs and examples of various augmented graphs obtained by applying various augmentation methods to the individual graphs.
- content accessed by a user in a corresponding session is sequentially shown instead of showing specific forms of individual graphs and augmented graphs.
- the 'selection' in Figure 3 is to select the augmented graph by selecting nodes A, C, and D, which are successive parts of individual graphs (F ⁇ A ⁇ C ⁇ D ⁇ B), and the lines between nodes A, C, and D. indicates that it can be obtained.
- 'Exclusion' in FIG. 3 indicates that an augmented graph can be obtained by excluding node C of an individual graph (F ⁇ A ⁇ C ⁇ D ⁇ B) and connecting a line between node A and node D.
- 'reorder' in FIG. 3 indicates that an augmented graph can be obtained by reversing the order of nodes A and C of individual graphs (F ⁇ A ⁇ C ⁇ D ⁇ B).
- the 'injection' in Figure 3 indicates that an augmented graph can be obtained by adding a new node C to an individual graph (F ⁇ A ⁇ C ⁇ D ⁇ B) and connecting a line between node B and the new node C.
- indicate 'Change' in FIG. 3 indicates that an augmented graph can be obtained by changing node B of an individual graph (F ⁇ A ⁇ C ⁇ D ⁇ B) to a new node E.
- nodes added to individual graphs may be determined based on the connection relationship and the number of connections of nodes connected to the integrated graph. For example, in the example of FIG. 3, if the previously obtained integrated graph is the integrated graph of FIG. 2, a new node C added in an augmented graph different from the injection method is connected to node C twice after node B of the integrated graph. It can be determined based on the fact that
- the connection relationship is not limited to a direct connection relationship. That is, in the example of FIG. 3, when the previously obtained integrated graph is the integrated graph of FIG. 2, a new node E added in an augmented graph different from the transition method is connected to node E as well as node D after node C of the integrated graph. It can be determined based on the fact that
- an individual graph can be augmented using only information in the individual graph, but it is difficult to obtain various augmented graphs because only information in the individual graph is considered. There is a limitation that it is difficult, and there may also be a problem that the information of the individual graph is lost during augmentation.
- an individual graph can be augmented using information included in an integrated graph in which information on a plurality of sessions is integrated as well as the corresponding individual graph, thus obtaining more types of augmented graphs. You can do it. And, as a result, by converting sequential data according to a plurality of sessions into a graph, it is possible to overcome the limitation that information transmission of nodes can occur only inside the session, and in particular, when the length of the session is small, various variations of the augmented graph ( variation) can be overcome.
- the processor 120 may obtain an augmented graph using the method of injecting and replacing as well as the method of selecting, excluding, and changing the order as described above.
- the processor 120 may randomly generate an augmented graph using the augmentation method as described above through the augmentation module, and at this time, the ratio of nodes that are changed in an individual graph according to the augmentation may be preset. there is.
- the augmentation module may be trained to obtain an augmented graph capable of reducing the contrast loss.
- the processor 120 can obtain an augmented graph having a distinct difference from an existing individual graph in a learning process while well preserving the characteristics of the individual graph through the augmentation module.
- the augmentation module may obtain three or more augmented graphs for each of the plurality of individual graphs by augmenting each of the plurality of individual graphs in three or more different methods, and thus, the contrast loss is increased by three or more enhancements.
- Two or more positive pairs defined based on vectors corresponding to the graph may be included, and as a result, more effective augmentation of individual graphs is possible.
- FIG. 4 is a block diagram showing the configuration of an electronic device 100 according to an embodiment of the present disclosure in detail.
- the electronic device 100 includes not only a memory 110 and a processor 120, but also a communication unit 130, an input interface 140, and an output interface 150. ) may further include.
- FIGS. 1 and 4 are merely examples, and various components other than those shown in FIGS. 1 and 4 may be included in the electronic device 100 , of course.
- the communication unit 130 includes a circuit and can communicate with an external device. Specifically, the processor 120 may receive various data or information from an external device connected through the communication unit 130 and may transmit various data or information to the external device.
- the communication unit 130 may include at least one of a WiFi module, a Bluetooth module, a wireless communication module, an NFC module, and a UWB module (Ultra Wide Band).
- each of the WiFi module and the Bluetooth module may perform communication using a WiFi method or a Bluetooth method.
- various connection information such as an SSID is first transmitted and received, and various information can be transmitted and received after communication is connected using this.
- the wireless communication module may perform communication according to various communication standards such as IEEE, Zigbee, 3rd Generation (3G), 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), and 5th Generation (5G).
- the NFC module may perform communication using a Near Field Communication (NFC) method using a 13.56 MHz band among various RF-ID frequency bands such as 135 kHz, 13.56 MHz, 433 MHz, 860 ⁇ 960 MHz, and 2.45 GHz.
- NFC Near Field Communication
- the UWB module can accurately measure Time of Arrival (ToA), which is the time at which a pulse arrives at a target, and Ange of Arrival (AoA), which is the angle of arrival of a pulse in a transmitting device, through communication between UWB antennas. It is possible to recognize precise distance and location within the error range of several tens of cm.
- ToA Time of Arrival
- AoA Ange of Arrival
- the processor 120 may receive information about content and information about a neural network model according to the present disclosure from an external device through the communication unit 130 .
- the processor 120 may control the communication unit 130 to transmit information on the user's access history to the external server, and through the communication unit 130 Information on recommended content may be received.
- the processor 120 may receive a user input for receiving information on the user's access history and recommended content information from the user terminal through the communication unit 130 .
- various information and data can be transmitted and received through the communication unit 130 in relation to external devices, external servers, and user terminals.
- the input interface 140 includes a circuit, and the processor 120 may receive a user command for controlling the operation of the electronic device 100 through the input interface 140 .
- the input interface 140 may include components such as a microphone, a camera (not shown), and a remote control signal receiver (not shown).
- the input interface 140 may be implemented as a touch screen included in a display.
- the microphone may receive a voice signal and convert the received voice signal into an electrical signal.
- the processor 120 may receive a user input corresponding to a user's access through the input interface 140, and receive a user input for receiving recommended content information.
- the processor 120 receives a user input, such as a user input for viewing content or a user input for loading a web page, through the input interface 140, and accesses the user based on the received user input.
- Information about history can be obtained.
- the output interface 150 includes a circuit, and the processor 120 may output various functions that the electronic device 100 can perform through the output interface 150 . Also, the output interface 150 may include at least one of a display, a speaker, and an indicator.
- the display may output image data under the control of the processor 120 .
- the display may output an image pre-stored in the memory 110 under the control of the processor 120 .
- the display according to an embodiment of the present disclosure may display a user interface stored in the memory 110 .
- the display may be implemented with a liquid crystal display panel (LCD), organic light emitting diodes (OLED), or the like, and the display may also be implemented with a flexible display, a transparent display, or the like, depending on circumstances.
- the display according to the present disclosure is not limited to a specific type.
- the speaker may output audio data under the control of the processor 120, and the indicator may be turned on under the control of the processor 120.
- the processor 120 may provide information on content and recommended content through the output interface 150 .
- the processor 120 may control the display to display content, and may also control the display to display information about recommended content while the content is displayed on the display.
- the processor 120 may display a user interface and receive a user input for receiving information on recommended content through the user interface.
- FIG. 5 is a diagram illustrating information on a user's access history in detail according to an embodiment of the present disclosure
- FIG. 6 illustrates an example of a user interface for providing information on recommended content based on the user's access history. it is a drawing
- the neural network model is a neural network model for acquiring information on recommended content based on a user's access history to video content.
- the information on the user's access history may include various pieces of information indicating that the user has accessed in the past or present with respect to various contents provided through the electronic device 100 .
- information about a user's access history may include information 510 for identifying the user.
- the information 510 for identifying a user may include information capable of identifying a user who accessed a plurality of contents in a specific session.
- the information 510 for identifying the user may include profile information about the user of the electronic device 100 or a user terminal connected to the electronic device 100 or login information about an application/website. .
- Information about the user's access history may include context information related to the user's access.
- the context information related to the user's access includes information 520 about the region where the user accessed the content, information 530 about the device the user used to access the content 530, and information about the time the user accessed the content. information 540.
- Information on the user's access history may include information 550 on content according to the user's access.
- the information 550 on content according to user access may include information capable of identifying content sequentially accessed by the user.
- an area 560 below the information 550 on content according to user access indicates that information on recommended content may be provided based on information on the user's access history.
- the processor 120 includes information 510 for identifying the user, information 520 on the region where the user accessed the content, information 530 on the device the user used to access the content, and information 530 on the user's access to the content. Based on the information 540 for one time and the information 550 for contents according to the user's access, the session according to the user's access can be classified, and the nodes of the individual graph corresponding to each divided session and the corresponding node It is possible to create individual graphs corresponding to each session by determining data corresponding to each session. That is, data corresponding to nodes of individual graphs may include not only information 550 on content according to user access, but also context information related to user access.
- the processor 120 may obtain information on recommended content corresponding to the individual graph using a neural network model learned according to the present disclosure and provide the information to the user. Also, information on recommended content may be provided through a user interface as shown in FIG. 6 .
- the processor 120 may control the display to display the first user interface 610 for receiving a user input.
- the first user interface 610 may include a UI item corresponding to a channel number as well as a UI item (“content recommendation” in FIG. 6 ) for receiving information on recommended content.
- the processor 120 displays a second user interface 620 including information on recommended content. You can control the display to show.
- information on recommended content may include thumbnail images of a plurality of recommended content.
- the processor 120 may control the display to display the selected one of the plurality of recommended contents.
- FIGS. 5 and 6 an example of a user interface for providing information on a user's access history and information on recommended content according to an embodiment of the present disclosure has been described with reference to FIGS. 5 and 6, an example as described above. It goes without saying that these are merely examples, and various other embodiments can be applied to the present disclosure. Another embodiment will be described with reference to FIGS. 7 and 8 .
- FIG. 7 is a diagram illustrating information on a user's access history in detail according to an embodiment of the present disclosure
- FIG. 8 illustrates an example of a user interface for providing information on recommended content based on the user's access history. it is a drawing
- the neural network model is a neural network model for obtaining information on a product that is recommended content based on a user's access history to a web page representing product information
- FIG. 7 a case in which products are shoes 710 , 720 , and 730 is illustrated as a premise.
- information on the user's access history includes information indicating that the user watched information on a specific product (View in FIG. 7), information indicating that the user added a specific product to the shopping cart (view in FIG. AddToCart), information indicating that the user has concluded a transaction for a specific product (Transaction in FIG. 7), and information indicating the time the user accessed the specific product (t 1 to t 7 in FIG. 7 ).
- the processor 120 includes information indicating that the user has viewed information on a specific product, information indicating that the user has added a specific product to the shopping cart, information indicating that the user has purchased a specific product, and information indicating that the user has accessed the specific product.
- Sessions according to user access can be classified based on information indicating time, and individual graphs corresponding to each session are created by determining nodes of individual graphs corresponding to each divided session and data corresponding to those nodes. can do.
- the processor 120 determines the blank interval between the time when the user made a transaction for a specific product (t 5 ) and the time the user watched information on the specific product thereafter (t 6 ) is a preset threshold. If the time is exceeded, access due to the user making a transaction for a specific product and access due to the user viewing information on the specific product thereafter may be divided into different sessions.
- the processor 120 selects a specific product rather than information indicating that the user has viewed information on a specific product.
- a higher weight may be assigned to information indicating that the user has added a specific product to the shopping cart, and a higher weight may be assigned to information indicating that the user has purchased a specific product rather than information indicating that the user has added a specific product to the shopping cart.
- the data corresponding to the nodes of the individual graph may include not only information indicating that the user has accessed a specific product, but also information about the user's access to the specific product.
- the processor 120 may obtain information on recommended content corresponding to the individual graph using a neural network model learned according to the present disclosure and provide the information to the user. Further, information on recommended content may be provided through a user interface as shown in FIG. 8 .
- the user interface may include recommended content information including an image of a product, a description of the product, and a price of the product, which are content recommended to the user.
- FIG. 9 is a flowchart illustrating a control method of the electronic device 100 according to an embodiment of the present disclosure.
- the processor of the electronic device 100 may obtain a plurality of individual graphs representing a user's access history for a plurality of contents for each of a plurality of sessions (S910). Specifically, the electronic device 100 may collect information about a user's access history to a plurality of contents for a predetermined period of time. In addition, the electronic device 100 may classify the user's access history for each of a plurality of sessions, and obtain a plurality of individual graphs representing the user's access history for each session.
- the electronic device 100 may generate an integrated graph in which the plurality of individual graphs are integrated based on the connection relationship and the number of connections of nodes included in the plurality of individual graphs (S920).
- the electronic device 100 may acquire a plurality of augmented graphs by augmenting each of a plurality of individual graphs based on the integrated graph (S930).
- the electronic device 100 may train a neural network model based on the plurality of augmented graphs. Specifically, the electronic device 100 may acquire contrast loss corresponding to a plurality of augmented graphs, and train a neural network model according to a contrastive learning method based on the obtained contrast loss (S940).
- control method of the electronic device 100 may be implemented as a program and provided to the electronic device 100 .
- a program including a control method of the electronic device 100 may be stored and provided in a non-transitory computer readable medium.
- the control method of the electronic device 100 records a user's access history for a plurality of contents in a plurality of sessions.
- the method may include acquiring a plurality of augmented graphs by augmenting each graph, and training a neural network model based on the plurality of augmented graphs.
- control method of the electronic device 100 and the computer readable recording medium including the program for executing the control method of the electronic device 100 have been briefly described, but this is only for omitting redundant description, and Various embodiments of the device 100 can also be applied to a computer readable recording medium including a control method of the electronic device 100 and a program executing the control method of the electronic device 100 .
- functions related to the neural network model as described above may be performed through a memory and a processor.
- a processor may consist of one or a plurality of processors. At this time, one or a plurality of processors are CPUs, general-purpose processors such as APs, GPUs. It may be a graphics-only processor, such as a VPU, or an artificial intelligence-only processor, such as an NPU.
- One or more processors control the input data to be processed according to predefined operating rules or artificial intelligence models stored in the non-volatile memory and the volatile memory.
- a predefined action rule or artificial intelligence model is characterized in that it is created through learning.
- being created through learning means that a predefined operation rule or an artificial intelligence model having desired characteristics is created by applying a learning algorithm to a plurality of learning data.
- Such learning may be performed in the device itself in which artificial intelligence according to the present disclosure is performed, or may be performed through a separate server/system.
- An artificial intelligence model may be composed of a plurality of neural network layers. Each layer has a plurality of weight values, and the layer operation is performed through the operation result of the previous layer and the plurality of weight values.
- Examples of neural networks include Convolutional Neural Network (CNN), Deep Neural Network (DNN), Recurrent Neural Network (RNN), Restricted Boltzmann Machine (RBM), Deep Belief Network (DBN), Bidirectional Recurrent Deep Neural Network (BRDNN), and GAN. (Generative Adversarial Networks) and deep Q-networks (Deep Q-Networks), and the neural network in the present disclosure is not limited to the above-described examples except for the cases specified.
- a learning algorithm is a method of training a predetermined target device (eg, a robot) using a plurality of learning data so that the predetermined target device can make a decision or make a prediction by itself.
- Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, and the learning algorithm in the present disclosure is specified Except for, it is not limited to the above example.
- the device-readable storage medium may be provided in the form of a non-transitory storage medium.
- 'non-temporary storage medium' only means that it is a tangible device and does not contain signals (e.g., electromagnetic waves), and this term refers to the case where data is stored semi-permanently in the storage medium and temporary It does not discriminate if it is saved as .
- a 'non-temporary storage medium' may include a buffer in which data is temporarily stored.
- the method according to various embodiments disclosed in this document may be included and provided in a computer program product.
- Computer program products may be traded between sellers and buyers as commodities.
- a computer program product is distributed in the form of a device-readable storage medium (eg compact disc read only memory (CD-ROM)), or through an application store (eg Play Store TM ) or between two user devices ( It can be distributed (eg downloaded or uploaded) online, directly between smartphones.
- a computer program product eg, a downloadable app
- a device-readable storage medium such as a manufacturer's server, an application store's server, or a relay server's memory. It can be temporarily stored or created temporarily.
- Each of the components may be composed of a single object or a plurality of entities, and some of the sub-components described above are omitted. or other sub-elements may be further included in various embodiments. Alternatively or additionally, some components (eg, modules or programs) may be integrated into one entity and perform the same or similar functions performed by each corresponding component prior to integration.
- operations performed by modules, programs, or other components may be executed sequentially, in parallel, repetitively, or heuristically, or at least some operations may be executed in a different order, may be omitted, or other operations may be added.
- unit or “module” used in the present disclosure includes units composed of hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic blocks, parts, or circuits, for example.
- a “unit” or “module” may be an integrated component or a minimum unit or part thereof that performs one or more functions.
- the module may be composed of an application-specific integrated circuit (ASIC).
- ASIC application-specific integrated circuit
- Various embodiments of the present disclosure may be implemented as software including commands stored in a storage medium readable by a machine (eg, a computer).
- the device calls the stored commands from the storage medium.
- a device capable of operating according to the called command it may include an electronic device (eg, the electronic device 100) according to the disclosed embodiments.
- the processor may directly or use other elements under the control of the processor to perform a function corresponding to the command.
- An instruction may include code generated or executed by a compiler or interpreter.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
Claims (15)
- 전자 장치에 있어서,신경망 모델을 저장하는 메모리; 및복수의 컨텐츠에 대한 사용자의 액세스 이력을 복수의 세션 별로 나타내는 복수의 개별 그래프를 획득하고,상기 복수의 개별 그래프에 포함된 노드들의 연결 관계 및 상기 노드들의 연결이 반복 되는 횟수에 기초하여 상기 복수의 개별 그래프가 통합된 통합 그래프를 생성하며,상기 통합 그래프에 기초하여 상기 복수의 개별 그래프 각각을 증강함으로써, 복수의 증강 그래프를 획득하고,상기 복수의 증강 그래프에 기초하여 추천 컨텐츠를 제공하기 위한 상기 신경망 모델을 학습시키는 프로세서; 를 포함하는 전자 장치.
- 제1 항에 있어서,상기 프로세서는,상기 복수의 개별 그래프 각각에 포함된 복수의 노드에 상기 통합 그래프에 포함된 적어도 하나의 노드를 추가함으로써, 상기 복수의 증강 그래프 중 적어도 하나의 증강 그래프를 획득하는 전자 장치.
- 제1 항에 있어서,상기 프로세서는,상기 복수의 개별 그래프 각각에 포함된 복수의 노드 중 적어도 하나의 노드를 상기 통합 그래프에 포함된 적어도 하나의 노드로 변경함으로써, 상기 복수의 증강 그래프 중 적어도 하나의 증강 그래프를 획득하는 전자 장치.
- 제1 항에 있어서,상기 프로세서는,상기 복수의 증강 그래프에 대한 대조 손실(contrastive loss)에 기초하여 상기 신경망 모델을 학습시키는 전자 장치.
- 제1 항에 있어서,상기 신경망 모델은,상기 복수의 개별 그래프에 대응되는 벡터들과 상기 복수의 증강 그래프에 대응되는 벡터들을 획득하는 인코더, 상기 복수의 개별 그래프 각각에 대응되는 추천 컨텐츠 정보를 획득하는 예측 모듈, 상기 복수의 개별 그래프 각각을 증강하여 상기 복수의 증강 그래프를 획득하는 증강 모듈 및 상기 복수의 증강 그래프에 대한 상기 대조 손실을 획득하는 대조 손실 획득 모듈을 포함하는 전자 장치.
- 제5 항에 있어서,상기 학습이 진행됨에 따라, 상기 인코더는 상기 복수의 증강 그래프 중 동일한 개별 그래프에 대응되는 증강 그래프들을 벡터 공간 상에서 가까운 거리의 벡터들로 매핑하고, 상기 복수의 증강 그래프 중 서로 다른 개별 그래프에 대응되는 증강 그래프들을 상기 벡터 공간 상에서 먼 거리의 벡터들로 매핑하도록 학습되는 전자 장치.
- 제5 항에 있어서,상기 복수의 증강 그래프에 대응되는 벡터들을 정의하기 위한 벡터 공간과 상기 복수의 증강 그래프에 대한 상기 대조 손실을 정의하기 위한 벡터 공간은 서로 상이한 전자 장치.
- 제1 항에 있어서,상기 증강 모듈은 상기 복수의 개별 그래프 각각을 서로 다른 세 가지 이상의 방법으로 증강하여 상기 복수의 개별 그래프 별로 세 개 이상의 증강 그래프를 획득하며,상기 대조 손실은 상기 세 개 이상의 증강 그래프에 대응되는 벡터들에 기초하여 정의되는 두 개 이상의 positive pair를 포함하는 전자 장치.
- 신경망 모델을 이용하는 전자 장치의 제어 방법에 있어서,복수의 컨텐츠에 대한 사용자의 액세스 이력을 복수의 세션 별로 나타내는 복수의 개별 그래프를 획득하는 단계;상기 복수의 개별 그래프에 포함된 노드들의 연결 관계 및 상기 노드들의 연결이 반복되는 횟수에 기초하여 상기 복수의 개별 그래프가 통합된 통합 그래프를 생성하는 단계;상기 통합 그래프에 기초하여 상기 복수의 개별 그래프 각각을 증강함으로써, 복수의 증강 그래프를 획득하는 단계; 및상기 복수의 증강 그래프에 기초하여 추천 컨텐츠를 제공하기 위한 상기 신경망 모델을 학습시키는 단계; 를 포함하는 전자 장치의 제어 방법.
- 제9 항에 있어서,상기 복수의 증강 그래프를 획득하는 단계는,상기 복수의 개별 그래프 각각에 포함된 복수의 노드에 상기 통합 그래프에 포함된 적어도 하나의 노드를 추가함으로써, 상기 복수의 증강 그래프 중 적어도 하나의 증강 그래프를 획득하는 단계; 를 포함하는 전자 장치의 제어 방법.
- 제9 항에 있어서,상기 복수의 증강 그래프를 획득하는 단계는,상기 복수의 개별 그래프 각각에 포함된 복수의 노드 중 적어도 하나의 노드를 상기 통합 그래프에 포함된 적어도 하나의 노드로 변경함으로써, 상기 복수의 증강 그래프 중 적어도 하나의 증강 그래프를 획득하는 단계; 를 포함하는 전자 장치의 제어 방법.
- 제9 항에 있어서,상기 신경망 모델을 학습시키는 단계는,상기 복수의 증강 그래프에 대한 대조 손실(contrastive loss)에 기초하여 상기 신경망 모델을 학습시키는 전자 장치의 제어 방법.
- 제9 항에 있어서,상기 신경망 모델은,상기 복수의 개별 그래프에 대응되는 벡터들과 상기 복수의 증강 그래프에 대응되는 벡터들을 획득하는 인코더, 상기 복수의 개별 그래프 각각에 대응되는 추천 컨텐츠 정보를 획득하는 예측 모듈, 상기 복수의 개별 그래프 각각을 증강하여 상기 복수의 증강 그래프를 획득하는 증강 모듈 및 상기 복수의 증강 그래프에 대한 상기 대조 손실을 획득하는 대조 손실 획득 모듈을 포함하는 전자 장치의 제어 방법.
- 제13 항에 있어서,상기 학습이 진행됨에 따라, 상기 인코더는 상기 복수의 증강 그래프 중 동일한 개별 그래프에 대응되는 증강 그래프들을 벡터 공간 상에서 가까운 거리의 벡터들로 매핑하고, 상기 복수의 증강 그래프 중 서로 다른 개별 그래프에 대응되는 증강 그래프들을 상기 벡터 공간 상에서 먼 거리의 벡터들로 매핑하도록 학습되는 전자 장치의 제어 방법.
- 제13 항에 있어서,상기 복수의 증강 그래프에 대응되는 벡터들을 정의하기 위한 벡터 공간과 상기 복수의 증강 그래프에 대한 상기 대조 손실을 정의하기 위한 벡터 공간은 서로 상이한 전자 장치의 제어 방법.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP22865040.4A EP4362477A4 (en) | 2021-08-31 | 2022-08-31 | ELECTRONIC DEVICE AND METHOD FOR CONTROLLING ELECTRONIC DEVICE |
| CN202280049821.7A CN117678227A (zh) | 2021-08-31 | 2022-08-31 | 电子装置和电子装置的控制方法 |
| US18/201,405 US20230297832A1 (en) | 2021-08-31 | 2023-05-24 | Electronic device and controlling method of electronic device |
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR20210115908 | 2021-08-31 | ||
| KR10-2021-0115908 | 2021-08-31 | ||
| KR1020220045188A KR20230032843A (ko) | 2021-08-31 | 2022-04-12 | 전자 장치 및 전자 장치의 제어 방법 |
| KR10-2022-0045188 | 2022-04-12 |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US18/201,405 Continuation US20230297832A1 (en) | 2021-08-31 | 2023-05-24 | Electronic device and controlling method of electronic device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023033538A1 true WO2023033538A1 (ko) | 2023-03-09 |
Family
ID=85411355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/KR2022/013026 Ceased WO2023033538A1 (ko) | 2021-08-31 | 2022-08-31 | 전자 장치 및 전자 장치의 제어 방법 |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US20230297832A1 (ko) |
| EP (1) | EP4362477A4 (ko) |
| WO (1) | WO2023033538A1 (ko) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117933304A (zh) * | 2024-02-08 | 2024-04-26 | 哈尔滨工业大学 | 一种基于自监督的多视角会话推荐桥接模型 |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12401853B2 (en) | 2022-09-28 | 2025-08-26 | Adeia Guides Inc. | Methods and systems for modifying a media guidance application based on user data |
| US12425692B2 (en) * | 2022-09-28 | 2025-09-23 | Adeia Guides Inc. | Methods and systems for modifying a media guidance application based on user data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20200040097A (ko) * | 2018-10-08 | 2020-04-17 | 삼성전자주식회사 | 전자 장치 및 그 제어 방법 |
| KR20200042739A (ko) * | 2018-10-16 | 2020-04-24 | 삼성전자주식회사 | 지식 그래프에 기초하여 콘텐트를 제공하는 시스템 및 방법 |
| KR20210028554A (ko) * | 2020-03-13 | 2021-03-12 | (주)빅인사이트 | 사용자의 시계열적 행동 데이터를 이용하여 인공지능 모델을 학습시키는 방법 및 장치 |
| KR20210028550A (ko) * | 2020-01-07 | 2021-03-12 | (주)빅인사이트 | 인공지능 학습 모델을 이용하여 사용자의 행동 데이터를 분석한 결과에 기초하여 사용자의 행동을 유도하는 방법 및 장치 |
| US20210081717A1 (en) * | 2018-05-18 | 2021-03-18 | Benevolentai Technology Limited | Graph neutral networks with attention |
-
2022
- 2022-08-31 EP EP22865040.4A patent/EP4362477A4/en active Pending
- 2022-08-31 WO PCT/KR2022/013026 patent/WO2023033538A1/ko not_active Ceased
-
2023
- 2023-05-24 US US18/201,405 patent/US20230297832A1/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210081717A1 (en) * | 2018-05-18 | 2021-03-18 | Benevolentai Technology Limited | Graph neutral networks with attention |
| KR20200040097A (ko) * | 2018-10-08 | 2020-04-17 | 삼성전자주식회사 | 전자 장치 및 그 제어 방법 |
| KR20200042739A (ko) * | 2018-10-16 | 2020-04-24 | 삼성전자주식회사 | 지식 그래프에 기초하여 콘텐트를 제공하는 시스템 및 방법 |
| KR20210028550A (ko) * | 2020-01-07 | 2021-03-12 | (주)빅인사이트 | 인공지능 학습 모델을 이용하여 사용자의 행동 데이터를 분석한 결과에 기초하여 사용자의 행동을 유도하는 방법 및 장치 |
| KR20210028554A (ko) * | 2020-03-13 | 2021-03-12 | (주)빅인사이트 | 사용자의 시계열적 행동 데이터를 이용하여 인공지능 모델을 학습시키는 방법 및 장치 |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP4362477A4 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117933304A (zh) * | 2024-02-08 | 2024-04-26 | 哈尔滨工业大学 | 一种基于自监督的多视角会话推荐桥接模型 |
Also Published As
| Publication number | Publication date |
|---|---|
| EP4362477A4 (en) | 2024-10-23 |
| US20230297832A1 (en) | 2023-09-21 |
| EP4362477A1 (en) | 2024-05-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023033538A1 (ko) | 전자 장치 및 전자 장치의 제어 방법 | |
| WO2018117704A1 (en) | Electronic apparatus and operation method thereof | |
| WO2019031714A1 (ko) | 객체를 인식하는 방법 및 장치 | |
| WO2016060480A1 (en) | Electronic device and method for spoken interaction thereof | |
| WO2021112631A1 (en) | Device, method, and program for enhancing output content through iterative generation | |
| EP3635605A1 (en) | Electronic device and method for controlling the electronic device | |
| EP3791328A1 (en) | Electronic device for reconstructing an artificial intelligence model and a control method thereof | |
| WO2018221824A1 (en) | Video communication device and video communication method | |
| EP3756145A1 (en) | Electronic apparatus and control method thereof | |
| WO2019245186A1 (ko) | 전자 장치 및 그의 제어 방법 | |
| WO2020105979A1 (en) | Image processing apparatus and control method thereof | |
| EP3539056A1 (en) | Electronic apparatus and operation method thereof | |
| WO2020040517A1 (en) | Electronic apparatus and control method thereof | |
| WO2020060151A1 (en) | System and method for providing voice assistant service | |
| WO2022197089A1 (ko) | 전자 장치 및 전자 장치의 제어 방법 | |
| WO2021040192A1 (ko) | 인공지능 모델을 학습시키는 시스템 및 방법 | |
| WO2020045794A1 (ko) | 전자 장치 및 그 제어 방법 | |
| WO2019190171A1 (ko) | 전자 장치 및 그 제어 방법 | |
| WO2020231243A1 (en) | Electronic device and method of controlling thereof | |
| WO2019231068A1 (en) | Electronic device and control method thereof | |
| WO2020116993A1 (ko) | 대화형 게임을 제공하는 전자 장치 및 그 동작 방법 | |
| WO2021071249A1 (en) | Electronic apparatus and control method thereof | |
| WO2023085736A1 (ko) | 전자 장치 및 전자 장치의 제어 방법 | |
| EP3738305A1 (en) | Electronic device and control method thereof | |
| WO2015122725A1 (en) | Method and apparatus for creating communication group |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22865040 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 202280049821.7 Country of ref document: CN |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2022865040 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2022865040 Country of ref document: EP Effective date: 20240124 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |