System and method for recommending virtual face

Abstract

The present invention relates to a system and a method for recommending a virtual face, wherein virtual face data is analyzed, the virtual face data is clustered on the basis of target information for which each virtual face is used, and when target information for which a virtual face is used is received by a user, a virtual face belonging to the target may be recommended. The present invention has an effect in enhancing the marketing index and enables recommendation of an ultra-personalized virtual face for each target for which a virtual face is used.

Description

Virtual face recommendation system and method

[0001] The present invention relates to a virtual face recommendation system and method, and more particularly, to a system and method for recommending a virtual model face that is effective in increasing marketing indicators.

[0002] Depending on the product type, target consumer type, or product concept, sellers seek models who can elicit positive customer responses. In these cases, sellers monitor model trends through media outlets like fashion weeks and magazines, or request model recommendations from modeling agencies.

[0003] In other words, in the past, we had no choice but to rely on limited model resources and unilaterally provided information without detailed classification according to market, product, or target consumer.

[0004] Moreover, depending on the target consumer and product type, which model would elicit a market response had to be determined by the intuition of experts, or the right or wrong model selection had to be judged based on sales results after the product was sold. Therefore, it was impossible to immediately identify and respond to the correlation between models and sales.

[0005] In addition, conventional recommendation systems typically provided recommendation services based on user ratings or review results for products or services, but as the number of user reviews increased, the amount of data to be processed increased exponentially, and conversely, there was also a cold-start problem in which the recommendation system did not work when there was little or no user review information.

[0006] Therefore, the technical problem that the present invention seeks to solve is to provide a technology that is effective in increasing marketing indicators and can recommend hyper-personalized virtual faces for each target for which the virtual face is used.

[0007] A virtual face recommendation system according to one aspect of the present invention for solving the above technical problem may include: a virtual face database storing virtual face data; a clustering module classifying the virtual face data based on target information for which each virtual face is used; a user data input module receiving target information for which the virtual face is used by a user; and a recommendation module recommending a virtual face belonging to a target selected by a user.

[0008] At this time, the target information may include one or more of a target industry, a target product, and a target customer.

[0009] Additionally, a verification module may be further included to verify whether marketing indicators increase after using the recommended virtual face.

[0010] In addition, the marketing indicators may include one or more of the visit rate, return visit rate, customer access time, page views, page stay time, purchase rate for the product, and repeat purchase rate for the web or mobile page.

[0011] In addition, the verification module can provide feedback so that the virtual face recommendation strength is set to increase or decrease based on the verification result through A / B testing.

[0012] At this time, the clustering module clusters virtual face data by target, and the recommendation module can receive virtual face data belonging to the cluster of targets selected by the user from the clustering module and recommend it.

[0013] In addition, the clustering module may include an encoder that extracts a virtual face feature vector from the virtual face data; and a clustering unit that clusters the virtual face feature vector and sets the target that occupies the largest proportion in each cluster as a target representing the cluster.

[0014] And, the above clustering unit can be clustered using a graph-based clustering algorithm.

[0015] Meanwhile, the system further includes an embedding layer; and a weighted sum layer, wherein the clustering module includes an encoder for extracting a virtual face feature vector from the virtual face data; and a clustering unit for classifying the virtual face data, wherein the user data input module includes a target input unit for receiving target information on how the virtual face is used by a user; and a usage history input unit for receiving virtual face usage history information of each user, wherein the embedding layer extracts target and usage history feature vectors from the target information and virtual face usage history information received from the target input unit and the usage history input unit, and the weighted sum layer assigns weights to the virtual face feature vector and the target and usage history feature vectors, respectively, and the recommendation module can recommend a virtual face as a result of a weighted sum for the virtual face feature vector and the target and usage history feature vectors, respectively.

[0016] At this time, the recommendation module can recommend a virtual face required by the user using NCF (Neural Collaborative Filter).

[0017] Meanwhile, a virtual face recommendation method according to one aspect of the present invention for solving the above technical problem may include a step of analyzing virtual face data as a method for recommending a virtual face by a system; a step of clustering the virtual face data based on target information on which each virtual face is used; and a step of recommending a virtual face belonging to a target when target information on which a virtual face is used by a user is received.

[0018] At this time, the target information may include one or more of a target industry, a target product, and a target customer.

[0019] In addition, after the above-mentioned recommending step, a step of verifying whether the marketing indicator increases after using the recommended virtual face may be further included.

[0020] At this time, the marketing indicator may include one or more of the visit rate, return visit rate, customer access time, page views, page stay time, purchase rate for the product, and repeat purchase rate for the web or mobile page.

[0021] And, after the above verification step, a step of increasing or decreasing the virtual face recommendation strength according to the verification result may be further included.

[0022] Meanwhile, the clustering step may be a step of clustering virtual face data by target, and the recommending step may be a step of recommending virtual face data belonging to a cluster of targets selected by the user.

[0023] And, the analyzing step may be a step of extracting a virtual face feature vector from virtual face data, and the clustering step may be a step of clustering the virtual face feature vector and setting the target with the largest proportion in each cluster as a target representing the cluster.

[0024] Additionally, the above virtual facial feature vector clustering may be a graph-based clustering algorithm.

[0025] Meanwhile, the analyzing step may include a step of extracting a virtual face feature vector from virtual face data, and the recommending step may include a step of receiving target information on which the virtual face is used by the user; a step of receiving virtual face usage history information of the user; a step of extracting a feature vector from the target information and the virtual face usage history information; and a step of assigning weights to the virtual face feature vector and the target and usage history feature vectors, respectively, and recommending a virtual face as a result of a weighted sum between the two vectors.

[0026] In addition, the above-mentioned recommended step may be a step of recommending a virtual face required by the user using NCF (Neural Collaborative Filter).

[0027] As described above, according to the present invention, it is effective in increasing marketing indicators and can recommend hyper-personalized virtual faces for each target for which the virtual face is used.

[0028] In addition, according to the present invention, the use of a virtual face model and its corresponding effects in the market can be immediately confirmed, and the confirmation results can be reflected back into model selection, making it more effective in increasing marketing indicators.

[0029] In addition, according to the present invention, since data such as reviews or ratings of other users are not used, the effect of reducing the amount of calculation for the recommendation algorithm can also be obtained.

[0030] Figure 1 is a diagram showing the overall configuration of a virtual face recommendation system according to one aspect of the present invention.

[0031] Figure 2 is a diagram showing the overall configuration of a virtual face recommendation system according to one aspect of the present invention.

[0032] Figure 3 is a diagram showing the overall configuration of a virtual face recommendation system according to one aspect of the present invention.

[0033] FIG. 4 is a flowchart illustrating an overall method for recommending a virtual face according to one aspect of the present invention.

[0034] Below, with reference to the attached drawings, embodiments of the present invention are described in detail so that those skilled in the art can easily implement the invention. However, the present invention can be implemented in various different forms and is not limited to the embodiments described herein. In addition, in the drawings, parts irrelevant to the description are omitted for clarity of description, and similar reference numerals are used throughout the specification to indicate similar parts.

[0035] Throughout the specification, whenever a part is said to "include" a component, this does not mean that it excludes other components, but rather that it may include other components, unless otherwise stated.

[0036] Additionally, terms such as “part,” “unit,” and “module” described in the specification mean a unit that processes at least one function or operation, which may be implemented by hardware, software, or a combination of hardware and software.

[0037] The devices described in the present invention are comprised of hardware including at least one processor, a memory device, a communication device, and the like, and a program that is executed by being combined with the hardware and stored in a designated location. The hardware has a configuration and performance capable of executing the method of the present invention. The program includes instructions that implement the operating method of the present invention described with reference to the drawings, and executes the present invention by being combined with hardware such as a processor and a memory device.

[0038] In this specification, “transmitting or providing” may include not only direct transmission or providing, but also indirect transmission or providing via another device or by using a bypass route.

[0039] In this specification, expressions described in the singular may be interpreted as singular or plural, unless explicit expressions such as “one” or “single” are used.

[0040] In this specification, the same drawing numbers refer to the same components regardless of the drawings, and “and / or” includes each and every combination of one or more of the mentioned components.

[0041] In this specification, terms including ordinal numbers, such as "first" and "second," may be used to describe various components, but these components are not limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present disclosure, a first component could be referred to as a "second component," and similarly, a second component could also be referred to as a "first component."

[0042] In the flowcharts described with reference to the drawings in this specification, the order of operations may be changed, several operations may be merged, some operations may be split, and certain operations may not be performed.

[0043] First, as illustrated in FIG. 1, a virtual face recommendation system (1) according to one aspect of the present invention includes a clustering module (10), a user data input module (20), a recommendation module (30), and a verification module (40), and can recommend a virtual face model capable of increasing marketing indicators.

[0044] In addition, the virtual face recommendation system (1) according to one aspect of the present invention may include a virtual face database (70).

[0045] The virtual face database (70) can store generated virtual face data and can store usage information for each virtual face. The usage information can include a label for the target for which each virtual face is used.

[0046] For example, the virtual face database (70) can store information about the industry / sales product / customer as information about the target for which each virtual face is used, for each of virtual faces A, B, and C.

[0047] The clustering module (10) can cluster virtual faces by target based on usage information about virtual faces stored in the virtual face database (70).

[0048] At this time, the target may include one or more of industry, product, and customer.

[0049] Accordingly, the clustering module (10) can cluster by industry, product, or customer, or, for example, by setting industry and product as major and minor classification items, respectively.

[0050] Alternatively, the clustering module (10) may cluster virtual faces according to the classification system by setting industries, products, and customers as major / medium / minor classification items, respectively.

[0051] The user data input module (20) can receive information about a target for which the user wishes to use the virtual face model.

[0052] Specifically, the user data input module (20) may receive information about the industry in which the virtual face model is used, or information about a product. Alternatively, the user data input module (20) may receive information about a customer in which the virtual face model is used.

[0053] The user data input module (20) may receive information solely for each industry, product, and customer, or may receive target information classified by major / medium categories or major / medium / subcategories, such as industry / product or industry / product / customer. In addition, the user data input module (20) may also receive target information, such as industry / customer or product / customer.

[0054] In addition, the user data input module (20) may present major / medium / minor categories respectively when the user inputs only one piece of information among industry, product, and customer, so that the user can input the remaining target information.

[0055] The recommendation module (30) can receive virtual face information belonging to a cluster from the clustering module (10) for a target selected by a user from the user data input module (20) and provide the information to the user by recommending it.

[0056] For example, if the recommendation module (30) receives target data for industry A / product B / customer C from the user data input module (20), it can check clustered virtual face data for the corresponding industry A / product B / customer C from the clustering module (10) and provide one or more virtual faces to the user.

[0057] The verification module (40) can verify the increased marketing indicator effect when using the recommended virtual face. Marketing indicators may include purchase rate, repeat purchase rate, visit rate for the relevant web page, revisit rate, user access time, page views, and page retention time.

[0058] For example, the verification module (40) can run an A / B test, and after using a virtual face, check the number of page views of the product or the company-related website to verify the degree of increase compared to the existing model face when using the virtual face.

[0059] The verification module (40) can feed back the verification result, for example, to the recommendation module (30), so that virtual faces with a negative marketing indicator increase effect can be excluded from the next recommendation or the recommendation strength can be set low.

[0060] Figure 2 is a configuration diagram of a virtual face recommendation system (1) according to one aspect of the present invention.

[0061] That is, the virtual face recommendation system (1) includes a clustering module (10), a user data input module (20), a recommendation module (30), a verification module (40), and a virtual face database (70), and can recommend a virtual face model that can increase marketing indicators.

[0062] Likewise, the virtual face database (70) can store generated virtual face data and usage information for each virtual face. The usage information can include a label for the target for which each virtual face is used.

[0063] The clustering module (10) can cluster virtual faces classified by target based on usage information for each virtual face, including an encoder (101) and a clustering unit (103). As described above, the target may include one or more of a target industry, a target product, or a target customer.

[0064] Specifically, the encoder (101) can encode a feature vector from virtual face data stored in a virtual face database (70).

[0065] For example, a deep learning-based face recognition algorithm as an encoder (101) can extract a feature vector from virtual face data.

[0066] The clustering unit (103) can cluster virtual face feature vectors for virtual faces classified by target.

[0067] At this time, for example, the clustering unit (103) can cluster using a graph-based clustering algorithm.

[0068] Graph-based clustering algorithms divide nodes within a graph into cohesive clusters based on common characteristics, with nodes assigned to the same cluster having more characteristics in common than nodes in other clusters.

[0069] Furthermore, graph-based clustering algorithms do not require assumptions about the shape or size of clusters. Therefore, they are useful in situations where the number of virtual faces to be clustered is unknown.

[0070] The clustering unit (103) can set the target with the largest proportion within each cluster as representing the cluster.

[0071] For example, according to the present invention, clusters a, b, and c can be formed by the clustering unit (103), and the target with the largest proportion in the a virtual facial feature vector cluster can be industry A / product A / customer A.

[0072] The user data input module (20) can receive information about a target for which the user wishes to use a virtual face model.

[0073] As described above, the user data input module (20) may receive information about the industry in which the virtual face model is used, or information about the product. Alternatively, the user data input module (20) may receive information about the customer in which the virtual face model is used.

[0074] The user data input module (20) may receive only information on each of industry, product, and customer, or may receive target information in the form of large / medium categories or large / medium / small categories, such as industry / product, industry / product / customer. In addition, the user data input module (20) may also receive target information, such as industry / customer, product / customer.

[0075] In addition, the user data input module (20) may present major / medium / minor categories respectively when the user inputs only one piece of information among industry, product, and customer, so that the user can input the remaining target information.

[0076] The recommendation module (30) can recommend and provide to the user a virtual face for a cluster corresponding to a target selected by the user among the targets representing each cluster.

[0077] For example, when clusters a, b, and c are formed by clustering units and the target represented in cluster b is industry B / product B / customer B, and the target selected by the user is industry B / product B / customer B, the recommendation module (30) can recommend a virtual face in the virtual face feature vector cluster b.

[0078] The verification module (40) can verify the increased marketing indicator effect when using the recommended virtual face. Marketing indicators may include purchase rate, repeat purchase rate, revisit rate, user access time, page views, and page retention time.

[0079] For example, the verification module (40) can run an A / B test, and as described above, by checking the page views of the website related to the product or the company after using the virtual face, it can verify the degree of increase compared to the existing model face when using the virtual face.

[0080] In addition, the verification module (40) can feed back the verification result, for example, to the recommendation module (30), so that virtual faces with a negative marketing index increase effect can be excluded from the next recommendation or have their recommendation priority set low.

[0081] Figure 3 is a configuration diagram of a virtual face recommendation system (1) according to one aspect of the present invention.

[0082] That is, the virtual face recommendation system (1) includes a clustering module (10), a user data input module (20), a recommendation module (30), a verification module (40), an embedding layer (50), a weighted sum layer (60), a virtual face database (70), and a user usage history database (80), and can recommend a virtual face model that can increase marketing indicators.

[0083] Likewise, the virtual face database (70) can store the generated virtual face database and can store usage information for each virtual face. In this case, the usage information can include a label for the target for which each virtual face is used.

[0084] The user usage history database (80) can store each user's virtual face usage history.

[0085] The clustering module (10) can cluster virtual faces classified by target based on usage information for each virtual face, including an encoder (101) and a clustering unit (103). As described above, the target may include one or more of a target industry, a target product, or a target customer.

[0086] Specifically, the encoder (101) can encode a feature vector from virtual face data stored in a virtual face database (70) and labeled for each target.

[0087] For example, as an encoder (101), a deep learning-based face recognition algorithm can extract a feature vector from a virtual face image.

[0088] The clustering unit (103) can cluster virtual faces classified by target, and a detailed description of the clustering unit (103) is as described above.

[0089] The user data input module (20) can receive user-related information, including a target input unit (201) and a usage record input unit (203).

[0090] Specifically, the target input unit (201) can receive information about a target for which the user wishes to use a virtual face model.

[0091] As described above, the target input unit (201) may receive information about the industry in which the virtual face model is used, or information about a product. Alternatively, the target input unit (201) may receive information about the customer in which the virtual face model is used.

[0092] The target input unit (201) may receive information solely for each industry, product, and customer, or may receive target information classified by major / medium category or major / medium / subcategory, such as industry / product or industry / product / customer. In addition, the target input unit (201) may also receive target information such as industry / customer or product / customer.

[0093] In addition, the target input section (201) may present major / medium / minor categories respectively when the user inputs only one piece of information among industry, product, and customer, so that the user can input the remaining target information.

[0094] The usage record input unit (203) can receive each user's virtual face usage record.

[0095] The embedding layer (50) can extract feature vectors from information about the user's target received from the target input unit (201) and the user's virtual face usage history received from the usage history input unit (203).

[0096] For example, the embedding layer (50) can feature vectorize information about a target input by a user through one-hot encoding.

[0097] The weighted sum layer (60) can assign set weights to each of the feature vectors extracted from the virtual face by the encoder (101) and the feature vectors received from the embedding layer (50).

[0098] For example, in the present invention, the recommendation module (30) described below can utilize NCF (Neural Collaborative Filter). Specifically, NCF (Neural Collaborative Filter) is a deep learning-based collaborative filtering algorithm that can predict items of interest to users based on the history of items with which each user interacted.

[0099] Therefore, in the present invention, the history of virtual faces interacted with by each user based on the virtual face usage history can be identified and utilized to recommend a virtual face for the user.

[0100] At this time, the weighted sum layer (60) sets a weight for the feature vector received from the embedding layer (50) related to the virtual face usage history of the user, so that it can be used as an input value when recommending a virtual face.

[0101] The recommendation module (30) can recommend a virtual face as a result of a weighted sum of a target-related virtual face feature vector and a user input target and virtual face usage history feature vector, according to the weights set by the weighted sum layer (60).

[0102] Therefore, according to the present invention, since the user's existing virtual face usage history is reflected and utilized for virtual face recommendation, a hyper-personalized virtual face can be recommended, thereby increasing user satisfaction.

[0103] The verification module (40) can verify the increased marketing indicator effect when using the recommended virtual face. Marketing indicators may include purchase rate, repeat purchase rate, revisit rate, user access time, page views, and page retention time.

[0104] For example, the verification module (40) can run an A / B test, and as described above, by checking the page views of the website related to the product or the company after using the virtual face, it can verify the degree of increase compared to the existing model face when using the virtual face.

[0105] In addition, the verification module (40) can feed back the verification result, for example, to the recommendation module (30), so that for virtual faces with a negative marketing index increase effect, the recommendation module (30) can add the loss value.

[0106] Conversely, the verification module (40) can set a high priority in the recommendation module (30) for virtual faces with a positive marketing indicator increase effect.

[0107] As described above, according to the present invention, the selection of a virtual face model and the marketing effect resulting from the selection are immediately confirmed, and it becomes easy to identify the correlation between the model and sales.

[0108] In addition, according to the present invention, new virtual faces can be recommended and provided flexibly according to market response, thereby improving the efficiency of marketing activities using models.

[0109] Referring to FIG. 4 below, a virtual face recommendation method according to one aspect of the present invention of the system (1) is summarized and explained.

[0110] The system (1) can first analyze the generated virtual face (S100, S101).

[0111] At this time, the system (1) may directly generate a virtual face or utilize the generated virtual face data.

[0112] The system (1) can classify virtual face data based on labeling for a target using a virtual face, and can also extract a feature vector from the virtual face data.

[0113] Next, the system (1) can cluster virtual faces by target (S103).

[0114] As described above, the target may include one or more of target industry, target product, and target customer, for example, by setting the industry / product / customer items as major / medium / minor categories, and clustering virtual faces according to the classification system.

[0115] Alternatively, the system (1) may perform target-specific clustering by clustering virtual facial feature vectors and setting the target with the largest proportion in each cluster as representing the cluster.

[0116] Meanwhile, the user can input target information for applying the virtual face model. For example, the system (1) can receive target industry / target product / target customer information to be used for marketing by applying the virtual face model (S105).

[0117] At this time, the system (1) can recommend a virtual face belonging to a cluster corresponding to the target input by the user (S107).

[0118] Specifically, the system (1) can recommend a virtual face belonging to a cluster corresponding to a user input target based on virtual face data clustered by target.

[0119] Alternatively, the system (1) can recommend a virtual face corresponding to a user input target based on a cluster according to virtual face feature vector clustering.

[0120] In addition, the system (1) may receive information on the target industry / target product / target customer, as information on the target who will use the virtual face, and may also receive information on the virtual face usage history of the corresponding user. In this case, the system (1) may also recommend a virtual face by referring to the virtual face feature vector and the virtual face usage history of the corresponding user (S105, S107).

[0121] For example, the system (1) may recommend a virtual face through a weighted sum between a virtual face feature vector belonging to the target and a feature vector for the user's virtual face usage history.

[0122] In addition, the system (1) can verify the effect of increasing marketing indicators when using the recommended virtual face (S109).

[0123] At this time, marketing metrics may include purchase rate, repeat purchase rate, return visit rate, user access time, page views, and page dwell time.

[0124] System (1) can reflect verification results, for example, from the virtual face analysis stage. System (1) can improve recommendation performance by reflecting verification result information, such as strengthening recommendations or lowering priorities for the virtual face.

[0125] The embodiments of the present disclosure described above are not implemented only through devices and methods, but may also be implemented through a program that realizes a function corresponding to the configuration of the embodiments of the present disclosure or a recording medium on which the program is recorded.

[0126] Although the embodiments of the present disclosure have been described in detail above, the scope of the present disclosure is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concepts of the present disclosure defined in the following claims also fall within the scope of the present disclosure.

Claims

1. A virtual face recommendation system, comprising: a virtual face database storing virtual face data; A clustering module that classifies the virtual face data based on target information used for each virtual face; A user data input module for receiving target information on which the virtual face is used by the user; and A recommendation module that receives target information from the user data input module and recommends a virtual face belonging to a target selected by the user; A system comprising 2. In paragraph 1, A system wherein the above target information includes one or more of a target industry, a target product, and a target customer.

3. In paragraph 2, A system further comprising a verification module for verifying whether marketing indicators increase after use of the recommended virtual face.

4. In paragraph 3, The above marketing indicators are: A system that includes at least one of the following: visit rate, return visit rate, customer access time, page views, time spent on page, purchase rate for the product, and repeat purchase rate for the web or mobile page.

5. In paragraph 4, The above verification module, A system that provides feedback so that the strength of the recommendation of the virtual face is set to increase or decrease based on the verification results through A / B testing.

6. In any one of paragraphs 1 to 5, The above clustering module clusters virtual face data by target, The above recommended module is, A system that receives virtual face data belonging to a cluster of targets selected by the user from the clustering module and recommends them.

7. In any one of paragraphs 1 to 5, The above clustering module, An encoder for extracting a virtual face feature vector from the above virtual face data; and A system comprising a clustering unit that clusters the above virtual facial feature vectors and sets the target with the largest proportion in each cluster as the target representing the cluster.

8. In paragraph 7, The above clustering unit is a system that clusters using a graph-based clustering algorithm.

9. In any one of paragraphs 1 to 5, Embedding layer; and Including additional weighted sum layers, The above clustering module, An encoder for extracting a virtual face feature vector from the above virtual face data; and Includes a clustering unit that classifies virtual face data, The above user data input module, A target input unit for receiving target information on which the virtual face is used by the user; and Including a usage history input section that receives information on each user's virtual face usage history; The above embedding layer extracts target and usage history feature vectors from target information and virtual face usage history information received from the target input unit and the usage history input unit, The above weighted sum layer assigns weights to each of the virtual face feature vector and the target and usage history feature vectors, The above recommendation module is a system that recommends a virtual face as a result of a weighted sum of the virtual face feature vector and the target and usage history feature vectors, respectively.

10. In paragraph 9, The above recommended module is, A system that recommends virtual faces that users need using NCF (Neural Collaborative Filter) 11. A method for the system to recommend a virtual face, Step of analyzing virtual face data; A step of clustering the virtual face data based on target information for which each virtual face is used; and A method comprising: when receiving target information in which a virtual face is used by a user, a step of recommending a virtual face belonging to the target; 12. In paragraph 11, The above target information is, A method comprising one or more of a target industry, a target product and a target customer.

13. In paragraph 12, After the above recommended steps, A method further comprising: a step of verifying whether a marketing indicator increases after using the recommended virtual face.

14. In paragraph 13, The above marketing indicators are: A method including at least one of the following: visit rate, return visit rate, customer access time, page views, page stay time, purchase rate for the product, and repeat purchase rate for the web or mobile page.

15. In paragraph 14, After the above verification step, A method further comprising the step of setting the virtual face recommendation strength upward or downward based on the verification result.

16. In any one of paragraphs 11 to 15, The above clustering step is, This is the step of clustering virtual face data by target. The above recommended steps are: A method for recommending virtual face data belonging to a cluster of targets selected by a user.

17. In any one of paragraphs 11 to 15, The above analysis steps are: This is a step of extracting a virtual face feature vector from virtual face data. The above clustering step is, A method for setting a target with the largest proportion in each cluster by clustering the above virtual facial feature vectors as a target representing the cluster.

18. In paragraph 17, The above virtual facial feature vector clustering method is a graph-based clustering algorithm.

19. In any one of paragraphs 11 to 15, The above analysis steps are: This is a step of extracting a virtual face feature vector from virtual face data. The above recommended steps are: A step of receiving target information whose virtual face is used by a user; A step of receiving information on the user's virtual face usage history; A step of extracting a feature vector from the above target information and virtual face usage history information; A method comprising a step of recommending a virtual face as a result of a weighted sum between the virtual face feature vector and the target and usage history feature vectors by assigning weights to the two vectors.

20. In paragraph 19, The above recommended step is a method for recommending a virtual face required by the user using NCF (Neural Collaborative Filter).

Images