KR102406634B1 - Personalized recommendation method and system based on future interaction prediction - Google Patents

Abstract

The present invention relates to a personalized recommendation method and system based on future interaction prediction, comprising: a memory for storing metadata of a user and content, and interaction information between the user and content; and accessing the memory, inputting metadata of the user and content, and interaction information between the user and content to learn a predictive model, and then calling a user embedding corresponding to a specific user ID, Estimate the embedding in time, predict the next interaction embedding of the specific user based on the estimated embedding at the current time of the specific user and the learned prediction model, and predict the next interaction embedding of the specific user and the learned and one or more processors for constructing a neural network that calculates the similarity between all content embeddings and returns a predetermined number of content similar to the predicted next interactive embedding of the specific user among all the learned content embeddings.

Description

PERSONALIZED RECOMMENDATION METHOD AND SYSTEM BASED ON FUTURE INTERACTION PREDICTION

The present invention relates to a content recommendation method, and more particularly, to a personalized recommendation method and system based on future interaction prediction.

As the speed of wired and wireless communication and the speed of development of operator resources are rapidly increasing, there is a huge amount of content online that can be said to be overflowing.

Accordingly, there is a great need for selecting and providing content preferred by a specific user from among such enormous content, and accordingly, various content recommendation models exist.

However, the conventional CF (Collaborative Filtering)-based recommendation and content-based recommendation technology simply recommends content according to a predefined rule.

The present invention uses deep learning-based interaction embedding techniques such as DeepCoevolve and JODIE to reflect the time and sequence context of the interaction between the user and the content, embedding the user and content in one vector space, and embedding the user and content vector, and predicting a content vector with which the user will interact next from the time information, and recommending content based on the similarity between the predicted vector and the content.

Domestic Registered Patent No. 10-2137887 Domestic Patent Publication No. 10-2013-0009360

The present specification has been devised to solve the above problems, and it is an object of the present specification to provide a future interaction prediction-based personalized recommendation method and system utilizing an embedding technique of deep learning-based interaction between a user and content. .

Another object of the present invention is to provide a personalized recommendation method and system based on future interaction prediction in which the time and sequence context of the interaction are reflected.

According to an embodiment of the present specification, in order to achieve the above object, according to an embodiment of the present specification, in the personalized recommendation method based on future interaction prediction according to the present specification, the personalized recommendation system includes the user and content metadata and the interaction between the user and content. learning the artificial neural network by inputting action information; retrieving, by the personalized recommendation system, a user embedding corresponding to a specific user ID; estimating, by the personalized recommendation system, embedding of a specific user at a current time; predicting, by the personalized recommendation system, the next interactive embedding of the specific user based on the estimated embedding at the current time of the specific user and the learned artificial neural network; calculating, by the personalized recommendation system, a degree of similarity between the predicted next interactive embedding of a specific user and all learned content embeddings; and returning, by the personalized recommendation system, a predetermined number of content similar to the predicted next interactive embedding of the specific user among all the learned content embeddings.

Preferably, the interaction information is characterized in that it indicates an interaction sequence consisting of an occurrence order and an occurrence time interval of the interaction between the user and the content.

Preferably, the metadata of the user includes at least one of ID, age, gender, address, and SNS record of the user.

Preferably, the metadata of the content includes an ID and a genre of the content.

Preferably, the step of training the artificial neural network comprises: processing, by the personalized recommendation system, metadata of the user and content, and interaction information between the user and content as input for embedding; embedding, by the personalized recommendation system, a user and content based on the user and content interaction information and metadata, and modeling a prediction of a next interaction; determining, by the personalized recommendation system, whether a ratio of the number of next interactions exceeds a reference value; and when the personalized recommendation system matches the prediction to the predicted total number of next interactions, and then the ratio of the number of interactions exceeds the reference value, the learned user and content embedding and the learned prediction model are stored It is characterized in that it comprises the step of

Preferably, in the modeling of the prediction of the next interaction, the personalized recommendation system models the prediction of the next interaction using any one of JODIE and DeepCoevolve.

Preferably, the step of invoking the user embedding comprises, by the personalized recommendation system, calculating a time difference between the previous interaction and the current time of the user.

Preferably, in the step of calculating the similarity, the personalized recommendation system uses any one of L2 Distance, Cosine Similarity, and Locality Sensitive Hashing (LSH) to embed the predicted next interaction of the specific user and the learning It is characterized by calculating the similarity between all content embeddings.

According to another embodiment of the present specification, a future interaction prediction-based personalized recommendation system according to the present specification includes: a memory for storing metadata of a user and content, and interaction information between the user and content; and accessing the memory, inputting metadata of the user and content, and interaction information between the user and content to learn a predictive model, and then calling a user embedding corresponding to a specific user ID, Estimate the embedding in time, predict the next interaction embedding of the specific user based on the estimated embedding at the current time of the specific user and the learned prediction model, and predict the next interaction embedding of the specific user and the learned and one or more processors for constructing a neural network that calculates the similarity between all content embeddings and returns a predetermined number of content similar to the predicted next interactive embedding of the specific user among all the learned content embeddings.

As described above, according to this specification, a personal recommendation method and system based on future interaction prediction using the weights of embeddings and prediction models learned through deep learning in which the history and time interval of the user's past interaction contents are provided. By doing so, it is possible to recommend content with high accuracy for each user.

1 is a block diagram illustrating an example of a function of a future interaction prediction-based personalized recommendation system according to an embodiment of the present invention;
2 is a block diagram showing a schematic configuration of the inside of the interactive embedding unit according to an embodiment of the present invention;
3 is a block diagram showing a schematic configuration of the inside of a recommendation unit according to an embodiment of the present invention;
4 is a block diagram showing an example of hardware capable of realizing the function of the personalized recommendation system according to the embodiment of the present invention;
5 is a flowchart showing a method for learning an artificial neural network according to an embodiment of the present invention;
6 is a flowchart illustrating a personalized recommendation method based on future interaction prediction according to an embodiment of the present invention;
7 is a diagram for explaining a comparison between the existing personalized recommendation method and the personalized recommendation method of the present invention;
8 is a diagram showing interaction information according to the present invention, and
9 is a diagram illustrating metadata according to the present invention.

It should be noted that the technical terms used herein are used only to describe specific embodiments, and are not intended to limit the present invention. In addition, the technical terms used in this specification should be interpreted as meanings generally understood by those of ordinary skill in the art to which the present invention belongs, unless otherwise specifically defined in this specification, and excessively inclusive It should not be construed in the meaning of a human being or in an excessively reduced meaning. In addition, when the technical terms used in the present specification are incorrect technical terms that do not accurately express the spirit of the present invention, they should be understood by being replaced with technical terms that can be correctly understood by those skilled in the art. In addition, the general terms used in the present invention should be interpreted according to the definition in the dictionary or according to the context before and after, and should not be interpreted in an excessively reduced meaning.

Also, as used herein, the singular expression includes the plural expression unless the context clearly dictates otherwise. In the present application, terms such as “consisting of” or “comprising” should not be construed as necessarily including all of the various components or various steps described in the specification, some of which components or some steps are It should be construed that it may not include, or may further include additional components or steps.

In addition, the suffixes "module" and "part" for components used in this specification are given or used in consideration of ease of writing the specification, and do not have a meaning or role distinct from each other by themselves.

Also, terms including ordinal numbers such as first, second, etc. used herein may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

In an embodiment of the present invention, the content may include webtoons, dramas, movies, animations, musicals, and the like.

In addition, in the embodiment of the present invention, 'embedding' represents expressing an object in a form that an artificial neural network can understand (ie, a vector).

Hereinafter, a preferred embodiment according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of reference numerals, and redundant description thereof will be omitted.

In addition, in the description of the present invention, if it is determined that a detailed description of a related known technology may obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention, and should not be construed as limiting the spirit of the present invention by the accompanying drawings.

1 is a block diagram illustrating an example of a function of a future interaction prediction-based personalized recommendation system according to an embodiment of the present invention.

Referring to FIG. 1 , the personalized recommendation system 100 according to the present invention may include an interactive embedding unit 110 and a recommendation unit 120 .

The interaction embedding unit 110 learns a predictive model by inputting user and content metadata and interaction information between the user and content. Here, the interaction information indicates an interaction sequence composed of an occurrence order and an occurrence time interval of interaction between the user and content. The user's metadata includes at least one of ID, age, gender, address, and SNS record of the user. The metadata of the content includes the ID and genre of the content.

The recommendation unit 120 retrieves the user embedding corresponding to the specific user ID, estimates the embedding at the current time of the specific user, and based on the estimated embedding at the current time of the specific user and the learned prediction model, the specific user predict the next interaction embedding of Returns the contents of

2 is a block diagram showing a schematic configuration of an inside of an interactive embedding unit according to an embodiment of the present invention.

Referring to FIG. 2 , the interaction embedding unit 110 according to the present invention includes a user-content interaction DB 210 , a user, content metadata DB 220 , a data loading unit 230 , and a metadata preprocessor ( 240 ), and a learning unit 250 .

The user-content interaction DB 210 stores the interaction between the user and the content. Here, the interaction may include viewing, purchasing, wishing, sharing, and view page information.

The user and content metadata DB 220 stores user and content metadata.

The data loading unit 230 generates interaction information in consideration of the interaction occurrence order and the occurrence time interval.

The metadata preprocessor 240 processes metadata of users and content as inputs for embedding. For example, the metadata preprocessor 240 converts metadata of users and content into vectors.

The learning unit 250 embeds the user and content based on the user and content interaction information and metadata, models the prediction of the next interaction, and when the prediction performance exceeds the reference value, the learned user and content embedding and , and store the learned predictive model. That is, the learning unit 250 matches the prediction of the predicted total number of next interactions, then determines whether the ratio of the number of interactions exceeds a reference value, and performs the prediction of the predicted total number of next interactions. When the ratio of the number of interactions after matching exceeds the reference value, the learned user and content embeddings and the learned predictive model may be stored.

3 is a block diagram showing a schematic configuration of the inside of a recommendation unit according to an embodiment of the present invention.

Referring to FIG. 3 , the recommendation unit 120 according to the present invention includes a time difference calculation unit 310 , a user projection unit 320 , a next content prediction unit 330 , a vector distance calculation unit 340 , and a sub-recommendation unit. (350).

The time difference calculator 310 calculates a time difference between the previous interaction and the user's current time. This is to train the artificial neural network differently when the user views the first content, the second content, and the third content in succession and then views the fourth content after a long time. That is, it is to exclude the fourth content from training the artificial neural network.

The user projection unit 320 calls a user embedding corresponding to a specific user ID.

The next content prediction unit 330 estimates the embedding of the specific user at the current time, and predicts the next interactive embedding of the specific user based on the estimated embedding at the current time of the specific user and the learned prediction model.

The vector distance calculator 340 calculates a similarity between the predicted next interactive embedding of a specific user and all the learned content embeddings. That is, the vector distance calculator 340 may calculate a vector distance between the predicted next interactive embedding of a specific user and all the learned content embeddings.

The lower recommendation unit 350 returns (recommends) a predetermined number of contents similar to the predicted next interactive embedding of a specific user among all the learned content embeddings. That is, the lower recommendation unit 350 returns the top K contents in which the vector distance between the predicted next interactive embedding of a specific user and all the learned content embeddings is less than a preset reference value.

Referring to FIG. 4 , hardware capable of realizing the functions of the personalized recommendation system 100 will be described. 4 is a block diagram showing an example of hardware capable of realizing the function of the personalized recommendation system according to the embodiment of the present invention.

The functions of the personalized recommendation system 100 can be realized using, for example, the hardware resources shown in FIG. 4 . That is, the function of the personalized recommendation system 100 is realized by controlling the hardware shown in Fig. 4 using a computer program.

As shown in FIG. 4 , this hardware mainly includes a CPU 402 , a ROM (Read Only Memory) 404 , a RAM 406 , a host bus 408 , and a bridge 410 . In addition, the hardware includes an external bus 412 , an interface 414 , an input unit 416 , an output unit 418 , a storage unit 420 , a drive 422 , a connection port 424 , and a communication unit 426 . has

The CPU 402 functions, for example, as an arithmetic processing unit or a control unit, based on various programs recorded in the ROM 404 , the RAM 406 , the storage unit 420 , or the removable recording medium 428 . Controls all or part of the operation of each component. The ROM 404 is an example of a storage device that stores a program read by the CPU 402, data used for arithmetic, and the like. In the RAM 406, for example, a program read by the CPU 402, various parameters that change when the program is executed, and the like are temporarily or permanently stored.

These elements are connected to each other via, for example, a host bus 408 capable of high-speed data transfer. On the other hand, the host bus 408 is connected to an external bus 412 having a relatively low data transfer rate, for example, via a bridge 410 . In addition, as the input part 416, a mouse, a keyboard, a touch panel, a touch pad, a button, a switch, a lever, etc. are used, for example. Also, as the input unit 416, a remote controller capable of transmitting a control signal using infrared or other radio waves may be used.

As the output unit 418 , for example, a display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), a plasma display panel (PDP), or an electro-luminescence display (ELD) may be used. Also, as the output unit 418, an audio output device such as a speaker or headphones, or a printer may be used.

The storage unit 420 is a device for storing various types of data. As the storage unit 420, for example, a magnetic storage device such as an HDD is used. As the storage unit 420 , a semiconductor storage device such as an SSD (Solid State Drive) or a RAM disk, an optical storage device, or a magneto-optical storage device may be used.

The drive 422 is a device that reads information recorded on the removable recording medium 428 that is a removable recording medium or writes information into the removable recording medium 428 . As the removable recording medium 428, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is used. Also, in the removable recording medium 428 , a program defining the operation of the personalized recommendation system 100 may be stored.

Connection port 424 is, for example, a USB (Universal Serial Bus) port, IEEE 1394 port, SCSI (Small Computer System Interface), RS-232C port, or an optical audio terminal, such as an external connection device 430 for connecting it's a port As the externally connected device 430, for example, a printer or the like is used.

The communication unit 426 is a communication device for connecting to the network 432 . As the communication unit 426, for example, a communication circuit for wired or wireless LAN, a communication circuit for WUSB (Wireless USB), a communication circuit for a cellular phone network, or the like can be used. The network 432 is, for example, a network connected by wire or wireless.

The hardware of the personalized recommendation system 100 has been described above. In addition, the above-mentioned hardware is an example, and a deformation|transformation which omits some elements, a deformation|transformation which adds new elements, etc. are possible.

5 is a flowchart illustrating a method for learning an artificial neural network according to an embodiment of the present invention.

Referring to FIG. 5 , the personalized recommendation system 100 processes metadata of a user and content as an input for embedding ( S510 ), and processes interaction information between the user and content as an input for embedding ( S520 ). Steps S510 and S520 may be performed simultaneously or sequentially.

The personalized recommendation system 100 embeds the user and the content based on the user and content interaction information and metadata, and models the prediction of the next interaction ( S530 ). In this case, the personalized recommendation system 100 may model the prediction of the next interaction using any one of JODIE and DeepCoevolve.

The personalized recommendation system 100 determines whether the prediction performance of the prediction model exceeds a reference value ( S540 ). Specifically, the personalized recommendation system 100 matches the prediction to the predicted total number of next interactions and then determines whether the ratio of the number of next interactions exceeds a reference value. The personalized recommendation system 100 returns to steps S510 and S520 when the prediction performance is less than or equal to the reference value.

The personalized recommendation system 100 stores the learned user and content embeddings when the prediction performance exceeds the reference value (ie, when the prediction model is trusted), and stores the learned prediction model ( S550 ).

6 is a flowchart illustrating a personalized recommendation method based on future interaction prediction according to an embodiment of the present invention.

The personalized recommendation method based on future interaction prediction according to the present invention recommends content to a user based on the prediction model learned in FIG. 5 above.

Referring to FIG. 6 , the personalized recommendation system 100 calls a user embedding corresponding to a specific user ID (S610), and calculates a time difference between the previous interaction and the user's current time (S620).

The personalized recommendation system 100 estimates embedding of a specific user at the current time (S630). At the same time, the personalized recommendation system 100 calls up the predictive model trained in FIG. 5 .

The personalized recommendation system 100 predicts the next interactive embedding of a specific user based on the estimated embedding at the current time of the specific user and the learned prediction model (S640). At the same time, the personalized recommendation system 100 calls all content embeddings previously learned in FIG. 5 .

The personalized recommendation system 100 calculates a similarity between the predicted next interactive embedding of a specific user and all the learned content embeddings (S650). That is, the personalized recommendation system 100 calculates the vector distance between the predicted next interaction embedding of a specific user and all the learned content embeddings. In this case, the personalized recommendation system 100 calculates the vector distance between the embedding of the next interaction of a specific user predicted using any one of L2 Distance, Cosine Similarity, and LSH (Locality Sensitive Hashing) and all the learned content embeddings. .

The personalized recommendation system 100 returns (recommends) a predetermined number of contents similar to the predicted next interactive embedding of a specific user among all the learned content embeddings (S660). That is, the personalized recommendation system 100 returns the top K contents in which the vector distance between the predicted next interactive embedding of a specific user and all the learned content embeddings is less than a preset reference value.

The above-described method may be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

In the case of implementation by hardware, the method according to embodiments of the present invention may include one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), and Programmable Logic Devices (PLDs). , FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers and microprocessors, and the like.

In the case of implementation by firmware or software, the method according to the embodiments of the present invention may be implemented in the form of a module, procedure, or function that performs the functions or operations described above. The software code may be stored in the memory unit and driven by the processor. The memory unit may be located inside or outside the processor, and may transmit/receive data to and from the processor by various well-known means.

7 is a diagram for explaining a comparison between the existing personalized recommendation method and the personalized recommendation method of the present invention.

Referring to FIG. 7A , since the existing personalized recommendation method does not consider the viewing order and time interval, the same result is recommended when users watch the same composition of the work.

Referring to FIG. 7B , since the personalized recommendation method according to the present invention reflects the viewing order and time interval, that is, the individual interaction sequence, different results are recommended even if users watch the same composition of the work.

8 is a diagram illustrating interaction information according to the present invention.

Referring to FIG. 8 , interaction information may include a user ID, a content ID, a timestamp, and an interaction type. Here, the interaction type may include viewing, purchasing, wishing, sharing, and view page information.

9 is a diagram illustrating metadata according to the present invention.

Referring to FIG. 9A , the user's metadata includes at least one of the user's ID, age, gender, address, and SNS record.

Referring to FIG. 9B , the metadata of the content includes the ID and genre of the content.

The embodiments disclosed herein have been described above with reference to the accompanying drawings. As such, the embodiments shown in each drawing should not be construed as being limited, and may be combined with each other by those skilled in the art after reading the content of the present specification, and when combined, it may be interpreted that some components may be omitted.

Here, the terms or words used in the present specification and claims should not be construed as being limited to conventional or dictionary meanings, but should be interpreted as meanings and concepts consistent with the technical idea disclosed in the present specification.

Therefore, the embodiments described in the present specification and the configurations shown in the drawings are only the embodiments disclosed in the present specification, and do not represent all the technical ideas disclosed in the present specification, so various equivalents that can be substituted for them at the time of the present application It should be understood that there may be water and variations.

110: interactive embedding unit 120: recommendation unit
210: user-content interaction DB 220: content metadata DB
230: data loading unit 240: metadata preprocessor
250: learning unit 310: time difference calculation unit
320: user projection unit 330: next content prediction unit
340: vector distance calculation unit 350: sub-recommended unit

Claims (9)

training, by the personalized recommendation system, an artificial neural network by inputting metadata of a user and content, and interaction information between the user and content;
retrieving, by the personalized recommendation system, a user embedding corresponding to a specific user ID;
estimating, by the personalized recommendation system, embedding of a specific user at a current time;
predicting, by the personalized recommendation system, the next interactive embedding of the specific user based on the estimated embedding at the current time of the specific user and the learned artificial neural network;
calculating, by the personalized recommendation system, a degree of similarity between the predicted next interactive embedding of a specific user and all learned content embeddings; and
returning, by the personalized recommendation system, a predetermined number of content similar to the predicted next interactive embedding of the specific user among all the learned content embeddings;
including,
The step of learning the artificial neural network is,
processing, by the personalized recommendation system, metadata of the user and content, and interaction information between the user and content as input for embedding;
embedding, by the personalized recommendation system, a user and content based on the user and content interaction information and metadata, and modeling a prediction of a next interaction;
determining, by the personalized recommendation system, whether a ratio of the number of next interactions after matching the prediction to the predicted total number of next interactions exceeds a reference value; and
When the personalized recommendation system matches the prediction of the predicted total number of next interactions, and the ratio of the number of interactions exceeds the reference value, the learned user and content embedding and the learned prediction model are stored step;
A personalized recommendation method based on future interaction prediction, comprising:
The method of claim 1,
The personalization recommendation method based on future interaction prediction, characterized in that the interaction information indicates an interaction sequence consisting of an occurrence sequence and an occurrence time interval of interaction between the user and the content.
The method of claim 1,
The user's metadata includes at least one of the user's ID, age, gender, address, and SNS record.
The method of claim 1,
The metadata of the content is a personalized recommendation method based on future interaction prediction, characterized in that it includes the ID and genre of the content.
delete The method of claim 1 , wherein modeling the prediction of the next interaction comprises:
The personalized recommendation system is a future interaction prediction-based personalized recommendation method, characterized in that the prediction of the next interaction is modeled using any one of JODIE and DeepCoevolve.
According to claim 1, wherein the step of calling the user embedding,
calculating, by the personalized recommendation system, a time difference between a previous interaction and the current time of the user;
A personalized recommendation method based on future interaction prediction, comprising:
According to claim 1, In the step of calculating the similarity,
The personalized recommendation system uses any one of L2 Distance, Cosine Similarity, and Locality Sensitive Hashing (LSH) to calculate the similarity between the predicted next interaction embedding of the specific user and all the learned content embeddings, characterized in that A personalized recommendation method based on future interaction prediction.
delete

Images