KR102563677B1 - The user interface method of providing cosmetic user profiling report based on user-written review - Google Patents

Abstract

The present invention relates to a user interface method for providing a cosmetic user profiling report based on user written reviews.
When a user terminal executes an application and requests generation of a cosmetic user profiling report, the service server generates a cosmetic user profiling report and presents it through the application screen of the user terminal. Through the profiling first section of the application screen, a predetermined number of representative topics of positive topic profiling information are visualized as data defining functional characteristics of the user's skin, and through the second section of the application screen, the user's skin trouble characteristics are visualized. As defining data, a predetermined number of representative topics of the negative topic profiling information are visualized.

Description

User interface method for providing cosmetic user profiling report based on user written review {THE USER INTERFACE METHOD OF PROVIDING COSMETIC USER PROFILING REPORT BASED ON USER-WRITTEN REVIEW}

The present invention relates to technology for providing personalized information to users using mobile applications.

Regarding cosmetics review, prior patent literature of the present applicant is Korean Patent No. 10-1955318. There are countless cosmetics in the world. Each cosmetic contains a wide variety of ingredients, and various functions are different for each cosmetic. People apply these cosmetics to their skin. People's desire for skin beauty and nutrition is great and strong. However, since people's skin characteristics are different, the efficacy of the same cosmetics varies from person to person. So, before buying cosmetics, people pay attention to not only its ingredients and functions, but also what other people's reputations are like, and what kind of people buy the cosmetics. Also, people have less hesitation in expressing their experiences and subjective evaluations of cosmetics. Because of this universal habit, review data on cosmetics has been explosively mass-produced through the Internet. Generally, such a review is written through a cosmetics information mobile application (ie, a cosmetics platform provided by a service server via the mobile application).

However, even though users contributed to the collection of big data by writing reviews on cosmetics, there were few benefits returned to users. The benefits of providing points were virtually everything. Despite expressing the experience of using cosmetics, there were still many users who could not accurately define the characteristics of their skin, and as a result, they repeatedly purchased products that did not suit their skin, and as a result, negative reviews were uploaded to the platform, which is necessary for the entity selling cosmetics. There was also a problem of spreading negative impressions.

The inventors of the present invention have conducted various researches and developments to find solutions for effectively solving these problems. And at the end of such efforts, the present invention has been completed.

An object of the present invention is to provide cosmetics user profiling information through which users can grasp at a glance skin characteristics that were difficult for them to define by analyzing reviews uploaded by users to a cosmetics platform. In this way, the user can be deeply aware of his or her own skin characteristics.

In addition, another object of the present invention is to write better quality responses and more reviews while reducing negative topics by inducing them to write reviews centered on their own skin characteristics. As a result, the platform secures more high-quality training data, and builds an environment in which better performance models can be created with such training data. Through the user's voluntary participation, the user understands their own skin characteristics more, and the business operator who wants to sell cosmetics has a better publicity effect, and as a result, a virtuous cycle structure that creates a cosmetics-related artificial intelligence model with better performance. want to establish

Meanwhile, other unspecified objects of the present invention will be additionally considered within the scope that can be easily inferred from the following detailed description and effects thereof.

The present invention for achieving the above object is a user interface method for providing a cosmetic user profiling report based on a user-written review:

When a user terminal executes a cosmetics information providing mobile application, accesses a cosmetics platform provided by a service server, and uploads N (N is an integer greater than 1) cosmetics reviews, the service server stores the cosmetics reviews in a cosmetics review database, , The topic classification server retrieves cosmetics reviews stored in the cosmetics review database, classifies topics, stores topic classification result values in the topic classification database, and the cosmetics user profiler uses the cosmetics reviews and the topic classification result values generate positive topic profiling information and negative topic profiling information;

When the user terminal requests generation of a cosmetics user profiling report, the service server generates a cosmetics user profiling report and presents it through the application screen of the user terminal, through the profiling first section of the application screen. Visualize a predetermined number of representative topics of the positive topic profiling information as data defining functional characteristics, and visualize a predetermined number of representative topics of the negative topic profiling information as data defining user skin trouble characteristics through the second section of the application screen. It is characterized in that it includes the step of visualizing the representative topic of the number.

In the user interface method for providing a cosmetic user profiling report based on a user-created review according to a preferred embodiment of the present invention, positive topic profiling information is pre-determined through the profiling third section of the application screen of the cosmetic user profiling report. It is preferable to further include a step of graphically presenting a representative topic selected from review data uploaded by a user terminal, including the number of representative topics, and the number of writings of the corresponding representative topic.

In addition, in the user interface method for providing a cosmetic user profiling report based on a user-created review according to a preferred embodiment of the present invention, the service server provides a representative topic included in the negative topic profiling information and a corresponding topic to the representative topic. It is preferable to further include a step of providing a review-component matching report in which cosmetic ingredient information is matched.

In addition, in the user interface method for providing a cosmetic user profiling report based on a user-created review according to a preferred embodiment of the present invention, the review uploaded by the user terminal through the profiling fifth section of the application screen of the cosmetic user profiling report. A step of classifying the topic list into a positive column and a negative column and providing the topic list may be further included.

Cosmetic reviews contain information such as performance, function, and effect of the product, and also include the user's skin condition and taste. Expressions containing such information can be converted from text to standardized data through a deep learning model. In addition, using standardized data, the present invention has the advantage of being able to profile which attributes the user considers important and what concerns the user has. Accordingly, there is an advantage in that the user can know his or her own skin characteristics or taste, thereby reducing the possibility of failing to purchase a product that does not suit him or her. In addition, the present invention exerts an effect of preventing unnecessary spread of bad reviews about products caused by ignorance of one's own skin.

On the other hand, even if the effects are not explicitly mentioned here, it is added that the effects described in the following specification expected by the technical features of the present invention and their provisional effects are treated as described in the specification of the present invention.

1 schematically shows the entire system configuration according to an embodiment of the present invention.
2 schematically shows an example of a module configuration of a profiling information generator 200 according to a preferred embodiment of the present invention.
Figure 3 schematically shows the entire process of a method according to a preferred embodiment of the present invention.
4 and 5 are screen scenarios as examples of visualizing a cosmetics user profiling report on the screen 100 of a cosmetics information providing mobile application according to a preferred embodiment of the present invention.
6 is a screen scenario as an example of visualizing a cosmetic user profiling report according to another preferred embodiment of the present invention.
* It is revealed that the accompanying drawings are illustrated as references for understanding the technical idea of the present invention, and thereby the scope of the present invention is not limited.

The present invention counts various cosmetics review texts of users as a deep learning-based model. This is because, after registering a specific word as a keyword in the system in advance, when such a corresponding keyword is found in the cosmetics review text, various expressions of the user can be reflected in the service more effectively than the method of counting the word. The rule-based methodology cannot properly reflect the various linguistic expressions of users who write cosmetic review texts. For example, by pre-registering the word 'applicability' as a rule base, if this word appears in the review text, it can be counted and analyzed. However, words or phrases that do not correspond to the rule base keywords cannot be analyzed. In that case, there are no words that match the keyword 'appliability', but semantically related reviews, such as 'balaem isn't very good' or 'there is no stickiness on the skin after application', should be classified separately . In addition, it is not possible to clearly present the conflicting meanings of a review with a complicated expression, 'Everyone says that the application is good, but I don't know', and a review that says, 'The application is good and it is moist, so it penetrates well into the skin'. The present invention implements a language model solution through prior training in the cosmetic review text by more cleverly improving the solution of the rule-based model for the cosmetic review text. In other words, it is possible to generate and present a more effective user profiling report by classifying not only the mention of 'appliability' but also the mention of 'good or bad application'.

Hereinafter, the principle and configuration of the present invention will be described as a preferred embodiment using the drawings. In the description of the present invention, if it is determined that a related known function may unnecessarily obscure the subject matter of the present invention as an obvious matter to those skilled in the art, the detailed description thereof will be omitted.

1 is a diagram schematically showing the system configuration of the present invention. The user terminal 1 is a computer device, preferably a smart phone. The cosmetics information providing mobile application 10 provided by the service server 2 is downloaded and installed. Regarding the basic configuration of the cosmetic information providing mobile application 10, reference may be made to descriptions of Korean Patent Nos. 10-1955318 and 10-2320003 of the present applicant.

In addition to the function of transmitting and receiving data by accessing the service server 2, the cosmetic information providing mobile application 10 of the present invention has a function of searching and viewing cosmetic information, writing and uploading reviews, and selecting and comparing cosmetics. , the function of purchasing and paying for cosmetics, the function of managing personal information, and the function of recommending cosmetics. In addition, these functions display information in data formats such as text, image, video, and voice, and it is structured as a user interface. In particular, the present invention includes a user interface related to a user profiling report.

The service server 2 is a server that provides the mobile application 10 to the user terminal 1 . The service server 2 that manages mobile applications and resources and software used by the applications is composed of one or more server devices, and the server devices include a plurality of hardware/software devices. It stores resources necessary for normal execution of applications, manages data, and performs communication functions. In addition, the present invention manages software that executes a user interface method that generates a user profiling report for each user and provides it through an application screen.

The service server 2 holds a plurality of databases. The cosmetics information database 25 may store cosmetics information to be provided to the user device through the mobile application of the present invention. Cosmetics information can be stored based on text/image/voice, and includes cosmetic ingredient information or functional information. The user information database 26 may record user information. In particular, user profile information such as the user's age, gender, skin type (dry, oily, neutral, combination), skin concerns (atopic, acne, sensitive skin), and whether or not there are children can be stored and managed for each user account.

The cosmetics review database 21 stores review data collected from users accessed through a network. The present invention is particularly related to a technique of processing and refining review data stored in the cosmetics review database 21 in several steps. In particular, in the present invention, this association is based on an artificial intelligence model of deep learning.

The topic classification server 20 reads the review text stored in the cosmetics review database 21 and classifies the topic. As a result of matching the review text with a pre-registered representative topic, the topic classification result value is stored in the topic classification database 23. In the figure, the cosmetics review database 21 and the topic classification database 23 are separated, but one database may be integrated and managed.

Preferably, the topic classification server 20 pre-generates at least one representative topic representing 'the function or characteristic of the cosmetic' among language expressions representing the function or characteristic of the cosmetic. The representative topic is a language expression corresponding to the interests of users who search for cosmetics information. For example, 'moisturizing', 'moisturizing', 'good application', 'non-stimulating', etc. can be representative topics.

In a preferred embodiment of the present invention, the service server 2 includes a cosmetics user profiler module. The cosmetics user profiler generates positive topic profiling information and negative topic profiling information using cosmetics reviews stored in the cosmetics review database 21 and topic classification result values stored in the topic classification database 23. What positive topic profiling information and negative topic profiling information are will be described later.

The cosmetic user profiler may include a representative topic generator and a profiling topic generator. The representative topic generator extracts cosmetics reviews that match a representative topic pre-registered to be displayed on the application screen from cosmetics reviews. The profiling information generator generates profiling information to be displayed as a cosmetics user profiling report using a language model previously determined in a corresponding cosmetics review.

2 schematically shows an example of a module configuration of a profiling information generator 200 according to a preferred embodiment of the present invention. For example, it includes a labeling module 210, a model building module 220, a model learning module 230, a model reasoning module 240, and a post-processing module 250. In addition, some texts of cosmetics reviews selected through step-by-step process processing of these modules can be visualized through profiling reports. Also, some sections of the review text may be highlighted and displayed. As in the example above, it is explained based on the topics of 'good application' and 'poor application'.

The labeling module 210 executes a labeling step of constructing labeling data for learning a topic classification model. From the large amount of cosmetic reviews built on the server through the labeling tool, reviews with 'good application' and 'poor application' are tagged. For a review that says, “The application is not very good,” it is tagged as “poor application.” Through this process, for example, 3,000 labeled text data is secured. In the labeled text, it is not simply expressed as 'not good for application', but also includes complexly expressed data such as 'everyone says that application is good, but I don't know'.

The model building module 220 executes a model building step for building a deep learning model. In a preferred embodiment of the present invention, a deep learning model of a language model, LSTM, bahdanau attention, and classifier structure is built.

First, let's explain the language model. The language model is a self-supervised learning model with a large amount of language corpus, and receives and uses the embedding value of the input text without updating the parameters of the model during learning. An embedding value is a value expressed as a number so that a computer can understand text data. LSTM (Long Short Term Memory) is a state in which the embedding value from the language model has not been learned in detail for 'correctness classification', so LSTM layers are additionally stacked to configure it to learn more about the task. A neural network model is created by stacking several neural network layers, and a neural network model in which a large number of neural network layers are stacked is called a deep learning model. In other words, by stacking additional LSTM layers, a deeper neural network structure is created.

Next, the Bahdanau Attention mechanism is applied. Finally, when 'correctness classification' is performed, the corresponding layer is additionally stacked to obtain the contribution of the segmental unit (this is called 'token') of the input text, that is, the attention weight value. The sum of the attention weights of tokens is 1. For example, the attention weight score for the sentence <The application is too poor> is as follows.

token Attention Weight spreadability 0.15 too 0.05 not 0.5 great 0.3

The topic classifier is a layer in which the result value for 'applicability classification' appears in the finally input text. That is, since the corresponding layer is implemented in a multiple label classification method, it is possible to determine whether the applicability is good, bad, or both. For example, a total of four classification results of [1,1], [1,0], [0,1], and [0,0] can be generated. This topic classifier can only classify two topics related to applicability, and in order to classify hundreds of topics, hundreds of classifiers are individually created and trained in the same way as above.

In other words, the topic classifier of the present invention has a multi-output structure with several classifiers instead of one classifier. Through this, the data imbalance problem and the relearning problem can be effectively solved. In the case of the conventional technique, if there is a text that reads <It is very easy to apply and moist and has no white cast, so it can be used very neatly>, it should be labeled as containing four topics: good application/moisture/no white cast/neat. In practice, labeling work in this way would require the labeler to read one text, look at hundreds of selection options, and select N corresponding ones. Virtually impossible and inefficient. Even if you have done that, if you need to add a new topic that you want to classify later, you have to look at tens of thousands of labeled texts again and label whether the topic is included. However, in the present invention, it is not configured like that. A topic classifier classifies N (N is an integer greater than 1) topics. The topic classifier is composed of multiple classifiers (language model + N classifiers) by including N individual classifiers that classify only the specific topic according to the language model for that topic. For one topic, a topic classifier composed of a language model and one individual classifier classifies the topic in a multi-label format. In other words, the topic 'applicability' is classified in a multi-label format of (good applicability, bad applicability). Accordingly, even when labeling the topic 'applicability', for example, it is only necessary to determine whether the applicability is good or bad in a given text. The same is true when adding a new topic.

Returning to FIG. 2 again, the model learning module 230 executes a model learning step of learning a language model using the topic classifier. Model training is not about training hundreds of classifiers at once. Each classifier is trained for the corresponding topic. Finally, the learned layer is individually attached to the language model in multiple ways so that it can be inferred at once. The model inference module 240 executes a model inference step of inferring a representative topic section for profiling from the input review text using the classification value of the representative topic received from the topic classifier and the contribution score received from the attention module. For the input text, the topic classifier receives a value for good or bad spreadability. It also receives a weight value for each token generated by the Attention layer. If it is classified as containing the content of 'poor applicability', it is possible to determine how much each token contributed to the classification through the attention weight. Next, the post-processing module 250 executes a post-processing step in which whether or not the section inferred in the model reasoning step is extended is reviewed and applied.

By the profiling information generator 200 of the present invention, not only the pre-registered references to 'applyability' are displayed, but also comments on 'good or bad applicationability' are classified so that the user's review text is effectively profiled. There are advantages to ringing.

3 schematically shows the entire process of a user interface method for providing a cosmetic user profiling report according to a preferred embodiment of the present invention.

First, the user accesses the cosmetics platform provided by the service server. Preferably, a cosmetics information providing mobile application is executed to access the cosmetics platform. Next, N (N is an integer greater than 1) cosmetics reviews are uploaded to the cosmetics platform (S100).

Then, the above-described topic classification server of the service server classifies the topic for the user's review text (S110). Preferably, topic information is classified, updated, and stored in user reviews at regular intervals. The cosmetic user profiler of the service server selects a pre-registered representative topic through the above topic classification. In addition, positive topic profiling information and negative topic profiling information are generated using the result values of cosmetics review and topic classification (S120). That is, corresponding information visualized to the user through the screen scenarios of FIGS. 4 to 6 of the present invention representing the user interface method is generated in advance by the service server. The positive topic profiling information may be specified as functionality considered important by the user, which is analyzed by the service server when the user writes a review on cosmetics posted on the cosmetics platform. Such information can be displayed as predefined representative topics, and these representative topics can be extracted by the modeling technique of FIG. 2 applied to the review text written by the user. For example, sticky, non-sticky, lasting, resilient, non-itchy, residue-free, waste removal, dandruff-free, dandruff-free, sebum-free, fixed, shiny, non-irritating, quick-drying, Good application, oily, light, rich, pore care, foaming, long lasting scent, moisturizing, refreshing, cleansing, non-drying, hair-improving, absorbing, moist, hydrating, spreadable, etc. It is a representative topic. Negative topic profiling information can be specified as the user's main skin concerns analyzed by the service server when the user writes a review on cosmetics posted on the cosmetics platform. Such information can be displayed as predefined representative topics, and these representative topics can be extracted by modeling techniques applied to review texts written by users. For example, representative topics such as improving hair quality, scent not lasting, long lasting, tightness, trouble, dryness, and poor application.

Next, the user terminal requests its profiling report to the service server (S130).

Then, the service server generates a cosmetic user profiling report and visualizes positive topic profiling information and negative topic profiling information through the application screen of the user terminal and presents them (S140).

4 illustrates a screen scenario of the user interface method of the present invention. FIG. 4(a) shows the first section 110 and the second section 120 of the application screen 100, and FIG. 4(b) shows the third section 130 of the application screen 100.

As shown in FIG. 4(a), in the profiling first section 110, a predetermined number of representative topics of positive topic profiling information are visualized as data defining functional characteristics of the user's skin. In the illustrated example, three representative topics, 'moist, moist, and absorbed', which are the representative topics counted the most as positive topics among the review data uploaded by the user, were visualized through the first section 110 of profiling. Next, in the second profiling section 120, a predetermined number of representative topics of negative topic profiling information are visualized as data defining the user's skin trouble characteristics. In the illustrated example, three representative topics such as 'dry, unhealthy hair, and trouble', which are the most counted representative topics as negative topics among the review data uploaded by the user, are visualized through the profiling second section 120. As shown in FIG. 4(b), in the third section 130, a representative topic selected from review data uploaded by a user terminal including a predetermined number of representative topics of positive topic profiling information and the corresponding representative topic The number of authors was presented graphically.

5 illustrates a screen scenario of the user interface method of the present invention. 5(a) shows the fourth sections 140 and 141 of the application screen 100, and FIG. 5(b) shows the fifth sections 150 and 151 of the application screen 100.

Among cosmetic review texts uploaded by users, negative topics have causality with specific ingredients of the corresponding cosmetic. Therefore, the service server of the present invention matches ingredient information stored in the cosmetics information database 25 with negative topics. That is, through the fourth sections 140 and 141 of the cosmetic user profiling report, a review-component matching report in which cosmetic ingredient information and negative topics are matched (ie, a representative topic of negative topics) is provided to the user. In the first area 140 of the fourth section, the message "This is a caution ingredient found in the product of the negative opinion written by (user)" is displayed, and in the second area 141 of the fourth section, one column Negative topics and cautionary ingredients of the cosmetic are displayed in correspondence with each other in the other column.

In addition, as shown in FIG. 5( b ), the review topic list uploaded by the user terminal is displayed through the profiling fifth sections 150 and 151 of the application screen 100 . Preferably, the positive column and the negative column are classified into independent regions and provided respectively. Convenience may be provided so that the review uploaded by the user can be viewed through the search interface 152 .

In the user interface method of the present invention, in some embodiments, the application screens of FIGS. 4 and 5 move up and down the screen of the smartphone through the user's input means (touch gesture input) as described above, and report their cosmetics user profiling. can be read. In another embodiment of the present invention, screens can be configured in layers, and information of each section can be displayed by pressing a button.

In addition, it is possible to access the cosmetic platform using a desktop or laptop as a user terminal, and in this case, a cosmetic user profiling report may be displayed as shown in FIG. 6 because the user screen is large. Even in this case, each section described above consists of an independent section of the user interface.

For reference, the user interface method for providing a cosmetic user profiling report based on user written reviews according to an embodiment of the present invention is implemented in the form of program commands that can be executed through various computer means and can be recorded on a computer readable medium. . The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in computer software.

Examples of computer readable media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and ROMs, RAMs, A hardware device specially configured to store and execute program instructions, such as flash memory, may be included. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine code such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The protection scope of the present invention is not limited to the description and expression of the embodiments explicitly described above. It is not limited by the visible shapes, figures, and proportions of the drawings provided for convenience of explanation. In addition, it is added once again that the scope of protection of the present invention cannot be limited due to obvious changes or substitutions in the technical field to which the present invention belongs.

Claims (4)

Uploading N (N is an integer greater than 1) cosmetics reviews by a user terminal by executing a cosmetics information providing mobile application to access a cosmetics platform provided by a service server that generates a cosmetics user profiling report for each user;
storing the cosmetics reviews in a cosmetics review database by a service server, and by a topic classification server loading cosmetics reviews stored in the cosmetics review database, classifying topics, and storing topic classification result values in a topic classification database;
generating, by a cosmetics user profiler, positive topic profiling information and negative topic profiling information in units of users using the cosmetics reviews and the topic classification result values;
requesting, by the user terminal, to generate a cosmetics user profiling report; and
A step in which the service server generates a cosmetics user profiling report for each user and presents it through the application screen of the user terminal, wherein the positive topic profile is data defining functional characteristics of the user's skin through the profiling first section of the application screen. Visualizing a predetermined number of representative topics of ring information and visualizing a predetermined number of representative topics of negative topic profiling information as data defining user skin trouble characteristics through a second section of the application screen , to correctly identify the characteristics of one's own skin,
In addition, a representative topic selected from the review data uploaded by the user terminal, including a predetermined number of representative topics of positive topic profiling information through the profiling third section of the application screen of the cosmetics user profiling report, and the corresponding representative topic A user interface method of providing a profiling report for a corresponding cosmetics user based on a user-written review, further comprising graphically presenting the number of writes.
delete According to claim 1,
Providing, by the service server, a review-component matching report in which a representative topic included in the negative topic profiling information and information on cosmetic ingredients corresponding to the representative topic are matched, the corresponding cosmetic product based on a review written by a user, further comprising: A user interface method that provides profiling reports for users.
According to claim 1,
Classifying the review topic list uploaded by the user terminal into a positive column and a negative column through the fifth section of the profiling section of the application screen of the cosmetics user profiling report, further comprising classifying and providing the review topic list to the corresponding cosmetics user based on the user-created review. Provide profiling reports for user interface methods.

Images