KR20230174986A - Apparatus and method for predicting favorite perfume - Google Patents

Abstract

Pertaining to a preferred perfume prediction device and method, the preferred perfume prediction device includes an input unit for obtaining target information and perfume information, extracting target keywords from the target information, generating target attribute vectors using the target keywords, and generating the perfume. It may include a processor that extracts a perfume keyword from information, generates a perfume attribute vector using the perfume keyword, and determines a preferred perfume using similarity between the subject attribute vector and the perfume attribute vector.

Description

Preferred perfume prediction device and method {APPARATUS AND METHOD FOR PREDICTING FAVORITE PERFUME}

It relates to a preference perfume prediction device and method.

Perfume is a cosmetic product made by diluting fragrances with alcohol and is known to have been used for approximately 5,000 years. Perfume can give olfactory pleasure by emitting scent to the outside. Recently, people have recognized perfume as a means of self-expression, and the demand for perfume that can reveal one's taste and personality is increasing. However, not only are there many types of perfume depending on the type of fragrance, mixing ratio, and dilution amount, but recently, indie perfumes containing the perfumer's personality and premium niche perfumes produced by professional perfumers to satisfy the tastes of a small number of people are also available. Nowadays, it is not easy to find a perfume that suits your taste and personality. In addition, unlike popular perfumes, indie perfumes and premium niche perfumes have a strong tendency for perfumers to explain the unique scent of the perfume by expressing it in poetic or abstract language. Such descriptions of perfumes are given by consumers, perfume sellers, or perfume recommenders. Because it requires the interpreter's active interpretation, there is a high possibility of interfering with the interpreter's subjectivity. Therefore, even if the perfume is purchased after reading the description, there are frequent cases where the perfume does not suit the consumer's taste. Accordingly, several services are being prepared to recommend perfumes that better suit consumer tastes, but these services have the inconvenience of having to continuously answer a large number of questions in the form of a tree to determine user preferences, and they have the disadvantage of having to continuously answer a large number of questions in the form of a tree to determine user preferences. If you want to get a recommendation for a perfume for yourself, there is a problem that you have to understand other people's tastes more accurately.

Republic of Korea Patent Publication No. 10-2022-0014131 (“Perfume recommendation device and method”, Konkuk University Global Industry-Academic Cooperation Foundation, publication date: 2022.02.04) Republic of Korea Patent Publication No. 10-2021-0117037 (“Method for recommending similar users in a social Internet of Things environment and recording medium recording the same”, Chungbuk National University Industry-Academic Cooperation Foundation, publication date: 2021.09.28)

A preferred perfume prediction device and a preferred perfume prediction method that can predict a suitable perfume for a person to whom a perfume is recommended by analyzing the information of the person to whom the perfume is recommended and text information about the perfume to detect perfume similar to the information about the person to whom the perfume is recommended. The problem to be solved is to provide .

In order to solve the above-mentioned problems, a preference perfume prediction device and a preference perfume prediction method are provided.

The preferred perfume prediction device includes an input unit for acquiring target information and perfume information, extracting target keywords from the target information, generating target attribute vectors using the target keywords, extracting perfume keywords from the perfume information, and extracting target keywords from the perfume information. It may include a processor that generates a perfume attribute vector using a keyword and determines a preferred perfume using the similarity between the subject attribute vector and the perfume attribute vector.

At least one of the subject information and the perfume information includes at least one sentence, and the processor performs morphological analysis on at least one of the subject information and the perfume information, removes stop words from the morphological analysis result, and At least one of the target keyword and the perfume keyword can be obtained.

When the target keyword is extracted, the processor detects a valid attribute keyword corresponding to a valid attribute among the target keywords, obtains a related keyword corresponding to the valid attribute keyword, and based on the target keyword and the related keyword The subject attribute vector can be generated.

When the perfume keyword is extracted, the processor counts the frequency of appearance of the perfume keyword and the number of documents in which the perfume keyword appears in at least one document of the perfume information, and determines the frequency of appearance of the perfume keyword in the document. You can determine the importance of words in a document by dividing the number.

The processor may obtain at least one core keyword from at least one perfume keyword based on word importance in the document and generate the perfume attribute vector using the at least one core keyword.

The processor obtains a similarity value by calculating Jaccard similarity between the subject attribute vector and the perfume attribute vector, and corresponds to a perfume attribute vector in which the similarity value is greater than a predefined reference value or the similarity value is relatively larger. You can also decide which perfume you like as your favorite perfume.

The processor may further utilize the subject's skin type to determine a preferred perfume for the subject, wherein the subject's skin type may include sensitive or oily.

The processor determines whether the perfume contains irritating chemical ingredients, whether a dark note is used, whether a note with a skin-soothing effect is included, whether a substance harmful to oily skin is included, whether a strong and long-lasting note is included, whether a moisturizing ingredient is included, and At least one additional alcohol content can also be used to determine the preferred perfume.

The preferred perfume prediction method includes obtaining target information and perfume information, extracting target keywords from the target information, generating target attribute vectors using the target keywords, and extracting perfume keywords from the perfume information. It may include generating a perfume attribute vector using the perfume keyword, and determining a preferred perfume using similarity between the subject attribute vector and the perfume attribute vector.

At least one of the target information and the perfume information may include at least one sentence, and the step of extracting target keywords from the target information includes performing morphological analysis on the target information and stopping words in the morphological analysis results. It may include the step of obtaining the target keyword by removing .

The step of generating an audience attribute vector using the audience keyword includes, when the audience keyword is extracted, detecting a valid attribute keyword corresponding to a valid attribute among the audience keywords, and obtaining a related keyword corresponding to the valid attribute keyword. and generating the target attribute vector based on the target keyword and the related keyword.

The step of generating a perfume attribute vector using the perfume keyword includes, when the perfume keyword is extracted, counting the frequency of appearance of the perfume keyword in at least one document of the perfume information, and counting the frequency of appearance of the perfume keyword in the document in which the perfume keyword appears. It may include counting the number and determining the importance of words in the document by dividing the number of documents by the frequency of appearance of the perfume keyword.

The step of generating a perfume attribute vector using the perfume keyword includes obtaining at least one core keyword from at least one perfume keyword based on word importance in the document, and generating the perfume attribute vector using the at least one core keyword. The step of generating an attribute vector may be further included.

The step of determining a preferred perfume using the similarity between the subject attribute vector and the perfume attribute vector includes obtaining a similarity value by calculating a Jacquard similarity between the subject attribute vector and the perfume attribute vector, and the similarity value is predefined. It may also include the step of determining a perfume corresponding to a perfume attribute vector that is greater than the reference value or has a similarity value that is relatively greater as the preferred perfume.

The step of determining a preferred perfume using the similarity between the subject attribute vector and the perfume attribute vector includes determining a preferred perfume for the subject using the similarity between the subject attribute vector and the perfume attribute vector and the subject's skin type. The step is further included, but the skin type of the subject may include sensitive or oily.

According to the above-described preferred perfume prediction device and preferred perfume prediction method, the perfume is selected by analyzing the information of the person to whom the perfume is recommended and text information about the perfume, and detecting perfumes similar to the information about the person to whom the perfume is recommended among a large number of perfumes. You can achieve the effect of being able to predict the appropriate perfume for the person to be recommended.

According to the above-described preferred perfume prediction device and preferred perfume prediction method, complex descriptions of perfumes are simplified through natural language analysis of the perfume description, and based on this, statistical-based analysis is performed to predict and recommend preferred perfumes with objective attributes. You can also get the effect of being able to

According to the above-described preferred perfume prediction device and preferred perfume prediction method, even when information about the person to whom the perfume will be recommended is limited, various attribute values for perfume recommendation can be built and configured through the process of extracting related keywords through a preference matching dictionary. As a result, it is possible to obtain the advantage of being able to derive a preferred perfume that is more suitable for the person to whom the perfume is recommended.

1 is a block diagram of an embodiment of a preferred perfume prediction device.
Figure 2 is a diagram for explaining an example of subject information and extracting keywords from subject information.
Figure 3 is a diagram for explaining an embodiment of a perfume database.
Figure 4 is a diagram for explaining an example of effective attribute information.
FIG. 5 is a diagram illustrating an example of valid attribute mapping and related keyword extraction for a specific user.
Figure 6 is a diagram for explaining an example of frequency for each word.
Figure 7 is a diagram showing an example of a description of a specific perfume.
Figure 8 is a diagram for explaining the number of specific word(s) and the number of documents in which the word appears (DF: Document Frequency).
Figure 9 is a diagram showing the result of multiplying the word appearance frequency (TF: Term Frequency) by the reciprocal of the number of documents in which the word appears.
Figure 10 is a diagram showing an example of an attribute table for a specific person.
Figure 11 is a diagram showing the results of determining the similarity, suitability for dry skin, suitability for oily skin, and suitability for sensitive skin of each perfume for a specific subject.
Figure 12 is a diagram for explaining an example of the process of determining whether an application is suitable for dry skin, oily skin, and sensitive skin.
Figure 13 is a flowchart of an embodiment of a preferred perfume prediction method.

Throughout the specification below, the same reference signs refer to the same components unless there are special circumstances. Terms with the addition of 'unit' used below may be implemented as software and/or hardware, and depending on the embodiment, one 'unit' may be implemented as one physical or logical part, or a plurality of 'units' may be implemented. It is also possible to be implemented with one physical or logical part, or one 'part' to be implemented with a plurality of physical or logical parts. When a part is said to be connected to another part throughout the specification, this may mean that the part and the other part are physically connected to each other and/or electrically connected. In addition, when a part includes another part, this does not mean excluding another part other than the other part unless specifically stated to the contrary, and means that another part may be included depending on the designer's choice. do. Expressions such as the first to Nth (N is a natural number greater than or equal to 1) are intended to distinguish at least one part(s) from other part(s), and do not necessarily mean that they are sequential unless otherwise specified. Additionally, singular expressions may include plural expressions, unless the context clearly makes an exception.

Hereinafter, an embodiment of a preferred perfume prediction device will be described with reference to FIGS. 1 to 12.

1 is a block diagram of an embodiment of a preferred perfume prediction device.

Referring to FIG. 1, in one embodiment, the preference perfume prediction device 100 may include an input unit 101, a storage unit 103, an output unit 109, and a processor 110. At least two of the input unit 101, storage unit 103, output unit 109, and processor 110 transmit electrical signals one-way or two-way through a circuit, cable, or wireless communication network to provide commands, data, etc. Can be transmitted. If necessary, at least one of the input unit 101, storage unit 103, and output unit 109 can be omitted.

The input unit 101 includes information 104 about a person who is the target of perfume recommendation (hereinafter referred to as target person), information about at least one perfume (105, hereinafter perfume information), or attributes valid for perfume recommendation. Information 106 (hereinafter referred to as valid attribute information) may be obtained and transmitted to the storage unit 103 or the processor 110. Additionally, the input unit 101 may receive a review of the preferred perfume from the user after the output unit 109 outputs the preferred perfume predicted by the processor 110. Reviews for the preferred perfume may be input from the user immediately or after a certain period of time has elapsed. The input unit 101 may include, for example, a keyboard device, a mouse device, a touch screen, a scanner device, an image capture module, a motion detection sensor, a pressure sensor, a proximity sensor, a microphone, a data input/output terminal, or a wired/wireless communication module (for example, a LAN It can be implemented using cards, short-range communication modules, mobile communication modules, etc.).

The storage unit 103 may temporarily or non-temporarily store at least one of the subject information 104, information about at least one perfume, and effective attribute information 106 input through the input unit 101. The storage unit 103 may include, for example, at least one of a main memory device and an auxiliary memory device.

Figure 2 is a diagram for explaining an example of subject information and extracting keywords from subject information.

The subject information 104 is information available for analysis of the subject, for example, information about the subject's individual and/or information about the perfume that the person operating the preferred perfume prediction device 100 (hereinafter referred to as the user) wants to search for. It may include information, etc. Here, the user and the target may be the same or different. In other words, the user can input information about a perfume that is appropriate for the user, or can input information about a perfume that is appropriate for someone other than the user (i.e., the target person) (for example, perfume for a gift). Information about the individual subject may include, for example, the subject's age, gender, or preferences. Information about perfume may include perfume manufacturing brand, perfume name, scent characteristics, scent intensity, notes, mood depending on the scent (mature, calm, or innocent, etc.), weather, region, and/or season, etc. Depending on the embodiment, the subject information 104 may include only at least one word or phrase, or may include at least one sentence(s) as shown in FIG. 2 . For example, the target information 104 is as shown in Figure 2: [Please recommend a perfume for men in their mid-20s. In this weather, which is better between XNell XLure Homme Sport and XAll XBage If you have any other good ones, please recommend them!], [Please recommend a perfume to give as a gift to a man in his mid-20s~ My favorite perfumes are X-Rosicos, X-Lari Light Essence, [Please recommend a women's perfume that is good to use in the summer~~ I've only used X All Depending on the embodiment, the subject information 104 may be obtained from one user, or may be obtained from two or more users (such as user A, user B, and user C).

According to one embodiment, the subject information 104 may be obtained by the user inputting an answer in response to a request that is output visually or auditorily through the output unit 109. For example, the output unit 109 first outputs the request to the outside in text, video, or audio format. The request may include a request to enter information about the subject's personal information or information about the perfume to be searched, and, if necessary, a request to enter information about the subject's skin type (oily, dry, or sensitive). It may also be included. In response to a request, the user may input subject information (104, words, sentences, or phrases, etc.) in the form of text, image (still image or video), or voice using a keyboard, mouse, or microphone.

Figure 3 is a diagram for explaining an embodiment of a perfume database.

Perfume information 105 is detailed information about at least one perfume(s) used to predict the subject's preferred perfume. For example, as shown in FIG. 3, the perfume information 105 includes at least one of the name of the perfume, a description of the perfume, notes of the perfume, and at least one user review (review) for the perfume. can do. If necessary, the perfume information 105 may include a series list of the corresponding perfume. Here, the description of the perfume is written by a perfumer, perfume maker, or perfume seller, and may be expressed literaryally using abstract or poetic language. A note refers to the scent of a perfume, for example each perfume has at least one note. The note may include at least one of a top note, a heart note, and a base note. Reviews are written directly by the subject, user, or others and may consist of symbols, numbers (scores, etc.), words, or sentences.

Perfume information 105 is entered directly into the preference perfume prediction device 100 by a user or administrator, or is entered into another terminal device (not shown) connected wired or wirelessly to the preference perfume prediction device 100 and is inputted into the preference perfume prediction device 100. It may have been obtained by passing it to (100). Depending on the embodiment, the perfume information 105 may be acquired from a server device (eg, a web server or a database server, etc.) accessible to the preference perfume prediction device 100 through a wired or wireless network. In this case, the perfume information 105 may have been collected through crawling or the like.

According to one embodiment, the perfume information 105 may be stored in the storage unit 103 in the form of a database (hereinafter referred to as perfume database). According to one embodiment, a perfume database may be constructed by assigning at least one identification number to a specific perfume and linking the perfume name, perfume description, usage notes, and at least one review, etc., corresponding to the identification number. Here, the perfume description, usage notes, and at least one review may be obtained through user or administrator input or website crawling. At least one review may also include a review of a preferred perfume recommended by the user from the preferred perfume prediction device 100.

Figure 4 is a diagram for explaining an example of effective attribute information.

The valid attribute information 106 may include at least one word, phrase, or sentence that can be used as a key attribute in predicting preference perfume. For example, as shown in Figure 4, valid attribute information includes words indicating age (10, 20, 30, 40, 50, early, mid, late, young or young, etc.), words indicating gender (male, female, male, female, masculine or feminine, etc.), words for seasons (such as spring, summer, fall or winter), words for weather (such as clear, cloudy, rain, snow or cool), name of perfume manufacturer, perfume brand Or it may include the perfume name, etc. This effective attribute information 106 may be predefined in advance. If necessary, the valid attribute information 106 may be updated by manual operation by a user or by control of the processor 110.

Additionally, the storage unit 103 may store at least one program (which may be referred to as an app, application, or software) that is driven to perform the operation of the processor 110. The program stored in the storage unit 103 may be written or modified directly by a designer such as a programmer and then stored in the storage unit 103, or may be stored in the storage unit 103 from another physical recording medium (external memory device, compact disk (CD), etc.). It may be delivered and stored, or it may be acquired or updated through an electronic software distribution network accessible through a wired or wireless communication network.

The output unit 109 may visually or audibly output an input request such as target information 104 or a preferred perfume prediction result of the processor 110 (for example, at least one preferred perfume or a list containing them) to the outside. there is. The output unit 109 may include, for example, a display, a printer device, a speaker device, an image output terminal, a data input/output terminal, or a communication module, but is not limited thereto. If necessary, the output unit 109 may be provided integrally with the input unit 101.

The processor 110 may perform a series of calculations required to predict a preferred perfume, obtain a preferred perfume prediction result, and then transmit the preferred perfume prediction result to the storage unit 103 or the output unit 109. In this case, the processor 110 may execute a program stored in the storage unit 103 to perform a preference perfume prediction operation. In addition, the processor 110 may automatically collect perfume information 105 according to a command input from a user or administrator or according to predefined settings and store it in the storage unit 103. Depending on the embodiment, the processor 110 may include a central processing unit (CPU), a graphics processing unit (GPU), a microcontroller unit (MCU), and an application processor (AP). It may include a processor, an electronic control unit (ECU), and/or at least one electronic device capable of performing various calculations and control processes. These processing or control devices may be implemented, for example, by using one or more semiconductor chips, circuits, or related components alone or in combination.

According to one embodiment, the processor 110 may include a subject analysis unit 120, a perfume analysis unit 130, and a preferred perfume prediction unit 140 as shown in FIG. 1, and depending on the embodiment, perfume information It may further include a collection unit 150. The subject analysis unit 120, the perfume analysis unit 130, the preferred perfume prediction unit 140, and the perfume information collection unit 150 may be logically or physically separated from each other, and depending on the embodiment, at least one of them may be One may be implemented using one or more physical processors (eg, a central processing unit and a graphics processing unit).

The subject analysis unit 120 analyzes the attributes of the subject based on the subject information 104, obtains a corresponding vector (hereinafter referred to as the subject attribute vector), and then sends the subject attribute vector to the preference perfume prediction unit 140. It can be delivered. In this case, the target analysis unit 120 may include a keyword extraction unit 121, an effective attribute determination unit 123, a related keyword acquisition unit 125, and a vectorization unit 127.

The keyword extraction unit 121 may extract at least one key word(s) 122 (hereinafter referred to as a target keyword) from the target information 104 as shown in FIG. 2 . For example, if the subject information 104 includes at least one sentence, the keyword extraction unit 121 performs morphological analysis on each sentence and, if necessary, removes stopwords from the morphological analysis results. Afterwards, at least one target keyword 122 corresponding to the target information 104 can be obtained. More specifically, for example, if user A, as shown in Figure 2, [Please recommend a perfume for a man in his mid-20s. In this weather, which is better between XNell XLure Homme Sport and XAll XBage If a number of sentences such as 'If there is anything else that is good, please recommend it!' are entered as the target information 104, the keyword extraction unit 121 performs morphological analysis of the target information 104 to extract at least one morpheme, and at least After excluding stop words such as perfume or recommendation from one morpheme, the remaining words or phrases (for example, man, Here, at least one target keyword 122 may have the form of a word, stem, or phrase. Stop words refer to meaningless or extremely small word tokens within a phrase or sentence, and may include words or phrases that are self-explanatory or difficult to use for analysis. Stop words may include words, stems, or phrases such as perfume, recommendation, these days, weather, -only, like, use, good, want, suitable, please, please, or please, but these are illustrative and user-friendly. I, the administrator, can define stop words in various ways depending on my choice. At least one extracted target keyword(s) may be delivered to the vectorization unit 127 and, depending on the embodiment, may also be delivered to the effective attribute determination unit 123. Additionally, operations such as morpheme analysis or stop word removal for each sentence may be performed using at least one learning model. Learning models may include, but are limited to, for example, a deep neural network (DNN), a convolutional neural network (CNN), and/or long short term memory (LSTM), etc. It doesn't work.

The valid attribute determination unit 123 receives at least one target keyword from the keyword extracting unit 121, and determines that the target keyword is valid attribute information 106 with reference to the valid attribute information 106 as shown in FIG. 4. You can decide whether it applies or not. For example, when the word [man] is transmitted as a target keyword, the valid attribute determination unit 123 may refer to the valid attribute information 106 and determine that [man] is a valid attribute. Additionally, if the word [maturity] is transmitted as the target keyword and the word is not included in the valid attribute information 160, the valid attribute determination unit 123 may determine that [maturity] is not a valid attribute. The valid attribute determination unit 123 detects and acquires at least one target keyword (hereinafter referred to as a valid attribute keyword) determined as a valid attribute according to the above-described process, and transmits the obtained effective attribute keyword to the related keyword acquisition unit 125. You can.

FIG. 5 is a diagram illustrating an example of valid attribute mapping and related keyword extraction for a specific user.

The related keyword acquisition unit 125 may receive a valid attribute keyword from the valid attribute determination unit 123 and obtain a related keyword 126 corresponding to the valid attribute keyword, as shown in FIG. 5 . The related keyword 126 may include at least one keyword related to a valid attribute keyword. According to one embodiment, the related keyword acquisition unit 125 may obtain one or more related keywords corresponding to the effective attribute keyword by mapping the effective attribute keyword to a predefined preference matching dictionary. More specifically, for example, when the related keyword acquisition unit 125 receives the word [mid-20s], which is a valid attribute keyword, it refers to the preference matching dictionary and provides at least one word corresponding to the received word [mid-20s]. It can be obtained by extracting related keywords, such as [bright], [abundance], [trend], [return to school], or [aqua]. The extracted related keywords 126 may be transmitted to the vectorization unit 127.

Figure 6 is a diagram for explaining an example of frequency for each word.

According to one embodiment, the vectorization unit 127 receives at least one target keyword from the keyword extraction unit 121 and simultaneously or sequentially receives at least one related keyword from the related keyword acquisition unit 125. , these can be combined to create a subject attribute vector. In this case, the vectorization unit 127 combines the target keyword(s) delivered by the keyword extraction unit 121 and the related keyword(s) corresponding to the effective attribute keywords to generate at least one set (hereinafter referred to as a keyword set), As shown in Figure 6, the frequency (for example, the number of times the same keyword is repeatedly extracted, etc.) is determined for each keyword belonging to the keyword set. For example, within a set of keywords, the keyword [aqua] has an appearance frequency of 7, the keyword [beach] has an appearance frequency of 6, and the keywords [white], [coast], and [summer] have an appearance frequency of 6. The frequency of appearance may be determined to be 5. Once the frequency corresponding to each keyword is determined, the vectorization unit 127 may extract one or more target attribute vector keywords from all keywords in the keyword set based on the frequency for each keyword. Depending on the embodiment, the vectorization unit 127 may extract N keywords as keywords for the target attribute vector in order of increasing frequency (where N is a natural number of 1 or more. Depending on the selection of the user or administrator, etc. can be determined), or at least one keyword whose corresponding frequency is greater than the standard frequency (for example, 3 or 5, etc. can be determined according to the selection of the user, administrator, etc.) (for example, the standard frequency is 4) If defined, [Aqua], [Beach], [White], [Coast], and [Summer], etc. in Figure 6) may be extracted as keywords for the subject attribute vector. Next, the vectorization unit 127 generates a target attribute vector by arranging the extracted keywords according to a predetermined order (for example, the frequency of each keyword). For example, the audience attribute vector for user A in Figure 2 is created as [aqua, beach, white, coastal, summer, light], and the audience attribute vector for user B is created as [gift, neutral, sea, reverberation, It can be created as [light]. The subject attribute vector may be transmitted to the preferred perfume prediction unit 140.

The perfume analysis unit 130 analyzes perfume properties based on perfume information (105, for example, a perfume database), acquires a corresponding vector (hereinafter referred to as a perfume attribute vector), and uses the obtained perfume attribute vector as a preferred perfume. It can be transmitted to the prediction unit 140. According to one embodiment, the perfume analysis unit 130 may include a keyword extraction unit 131 and a vectorization unit 133.

FIG. 7 is a diagram illustrating an example of a description of a specific perfume, and FIG. 8 is a diagram illustrating the number of each specific word(s) and the number of documents in which the word appears (DF: Document Frequency). Figure 9 is a diagram showing the result of multiplying the word appearance frequency (TF: Term Frequency) by the reciprocal of the number of documents in which the word appears.

The keyword extraction unit 131 may extract one or more keywords (hereinafter referred to as perfume keywords) from the perfume information 105. For example, when the perfume information 105 includes at least one sentence, the keyword extractor 131 performs morphological analysis on each of the at least one sentence to obtain at least one word or phrase, and Accordingly, one or more perfume keywords corresponding to the perfume information 105 can be extracted from the perfume information 105 by removing stop words from the morphological analysis results (i.e., words, phrases, etc.). For example, as shown in Figure 7, if there are multiple documents (D0 to D3, each document may include at least one sentence about perfume) regarding a specific perfume (for example, perfume A), the keyword The extraction unit 131 performs morphological analysis on each of these documents (D0 to D3) to obtain a number of words or phrases, and among the acquired words or phrases, stop words (for example, [perfume] or [taste] [Sniping], etc.) to remove [Wind], [Coast], [White], [Sand], and [Summer] from all documents (D0 to D3). You can obtain perfume keywords such as [Woody], [Teen], [Women], [Grapefruit], [Good], and [Beach]. Morphological analysis or stop word removal for a sentence may be performed using at least one learning model.

When the perfume keyword is extracted, the keyword extraction unit 131 may determine the word appearance frequency (TF) based on each document (D0 to D3), as shown in FIG. 8. The word appearance frequency (TF) refers to the number of times a certain word (term) is detected in one document (D0 to D3). Specifically, for example, the keyword extraction unit 131 determines that at least one word (e.g., [wind], [coast], or [daily life]) in a specific document (D0 to D3) for a specific perfume, respectively. By counting the frequency with which each word appears, the word appearance frequency (TF) corresponding to each word can be obtained and recorded. For example, as shown in Figure 8, in the case of document 0 (D0), [Wind] is used once, [Coast] is used twice, [Daily Life] is used twice, [Cool] is used once, [ It can be determined that [Woody] appeared 1 time, [Beach] 0 times, [Summer] 3 times, [Woman] 0 times, and [Sand] 0 times.

Additionally, the keyword extraction unit 131 may further count the number of documents (DF) in which a word appears for a specific keyword. The number of documents in which a word appears (DF) means the number of documents (D0 to D3) in which a given word appears among all documents (D0 to D3). Specifically, referring to Figure 8, the keyword [wind] appears in two documents (D0 and D2), so the number of documents (i.e., the number of documents (DF) in which the word appears) corresponding to the keyword [wind] The number becomes 2. Likewise, since the keyword [coast] appears in two documents (D0 and D2), the number of documents (DF) in which the word corresponding to the keyword [coast] appears is also 2. Likewise, the number of word-appearing documents (DF) for each keyword [daily life], [cool], [woody], [beach], [summer], [woman], and [sand] is 3, 1, 1, respectively. Can be given as 2, 2, 2 and 1.

When the word appearance frequency (TF) and the number of word appearance documents (DF) are determined based on each keyword and document (D0 to D3), the keyword extraction unit 131 determines the word for each keyword as shown in FIG. 9. Based on the frequency of appearance (TF) and the number of documents in which the word appears (DF), the importance of the word in the document can be determined. According to one embodiment, when at least one word is given, the keyword extractor 131 divides the word appearance frequency (TF) by the number of word appearance documents (DF) for the word, and divides the resulting value (TF/DF). Based on this, the importance of words in a document can be determined. For example, referring to Figures 8 and 9, the importance within document 0 (D0) for the keyword [wind] is the word appearance frequency of the keyword [wind] for document 0 (D0). Since (TF) is 1 and the number of documents in which the word appears (DF) is 2, it is given as 0.5 (=1/2). Likewise, the importance of the keyword [wind] in document 1 (D1) has a value of 0 because the word appearance frequency (TF) of the keyword [wind] in document 0 (D0) is 0. In addition, the importance of the keyword [coast] in document 0 (D0) is that the word appearance frequency (TF) of the keyword [coast] for document 0 (D0) is 2 and the number of documents in which the word appears (DF) is 2. Therefore, it is decided as 1 (=2/2). As shown in Figure 9, the obtained word importance within the document is transmitted to the vectorization unit 133.

The vectorization unit 133 may select core keywords from the extracted perfume keywords and generate a perfume attribute vector based on the selected core keywords. For example, when the importance of words in a document is transmitted as shown in FIG. 9, the vectorization unit 133 selects at least one nostalgia keyword (for example, [wind], [coast], [daily life], select at least one of [cool], [woody], [beach], [summer], [feminine], and [sand], etc.), determine at least one perfume keyword selected as a core keyword, and determine at least one core keyword selected You can create a perfume attribute vector using keywords. According to one embodiment, the vectorization unit 133 may determine and select at least one perfume keyword as a core keyword based on word importance (TF/DF) in the document. In this case, the vectorization unit 183 may extract M keywords with high word importance (TF/DF) in the document (M is a natural number of 1 or more. It may be predefined by the user or administrator) as core keywords. And/or keywords with a word importance (TF/DF) value within the document greater than a predefined standard value may be extracted as core keywords. When at least one core keyword is obtained, the vectorization unit 133 may generate at least one perfume attribute vector by arranging the at least one core keyword according to a predefined definition. Depending on the embodiment, the vectorization unit 133 counts the frequency of each key keyword for all or some documents (D0 to D3), and according to the counting result, the frequency is relatively high or exceeds a predetermined standard value. A perfume attribute vector may be generated by extracting keywords (hereinafter referred to as additional extracted keywords) from core keywords and arranging the additional extracted keywords according to predefined criteria. Here, the predefined criteria may include, for example, the frequency of additional extracted keywords. Accordingly, a perfume attribute vector corresponding to each perfume can be obtained. For example, the perfume attribute vector for perfume 1 is [coastal, everyday, beach, summer, woman…] ], and the perfume attribute vector for perfume 2 is [warm, floral, intense, fruity, romantic... ] can be determined. The perfume attribute vector may be transmitted to the preferred perfume prediction unit 140.

The preferred perfume prediction unit 140 may receive the subject attribute vector and the perfume attribute vector, and predict the preferred perfume using the subject attribute vector and the perfume attribute vector. According to one embodiment, the preferred perfume prediction unit 140 may calculate and determine the similarity between the subject attribute vector and the perfume attribute vector, and determine at least one preferred perfume for the subject based on the calculated similarity.

FIG. 10 is a diagram showing an example of an attribute table for a specific subject, and FIG. 11 is a diagram showing the results of determining whether each perfume is suitable for dry skin, oily skin, and sensitive skin.

According to one embodiment, the preferred perfume prediction unit 140 first performs embedding by applying one hot encoding to the subject attribute vector and the perfume attribute vector, and in response, as shown in FIG. 10, the subject attribute Table 151 can be created. One-hot encoding creates a vector consisting of multiple elements, and vector creation can be performed by adding 1 to elements containing a specific keyword and adding 0 to other elements. For example, the subject attribute table 151 sets the element value for that word to 1 if a specific word (for example, beach or coastal, etc.) exists in the perfume attribute vector for a specific perfume, and sets it to 0 otherwise. Thus, a value of 0 or 1 is recorded in the element for all words, and similarly, if a specific word exists in the subject attribute vector for a specific subject, the element value for the specific word is defined as 1, otherwise it is defined as 0. It could be. The created subject attribute table 151 can be used for similarity calculation.

According to one embodiment, the preferred perfume prediction unit 140 may calculate the similarity between the subject attribute vector and the perfume attribute vector using Jaccard similarity. Jacquard similarity is used to measure the similarity between two sets (A, B). As described in Equation 1 below, the size of the intersection of at least two sets (A, B) is calculated as the union of at least two sets (A, B). It can be calculated by dividing by the size of . In this case, the closer the Jacquard similarity calculation result is to 1, the more similar the two sets are, and conversely, the closer the Jacquard similarity calculation result is to 0, the less similar the two sets are.

[Equation 1]

Here, each set (A, B) includes a subject attribute vector and a perfume attribute vector.

According to the Jaccard similarity operation between the perfume attribute vector for each perfume and the subject attribute vector for the subject, at least one similarity value corresponding to each perfume can be obtained, as shown in FIG. 11. For example, the similarity between the perfume attribute vector for perfume 1 and the subject attribute vector is calculated as 0.68, the similarity between the perfume attribute vector for perfume 2 and the subject attribute vector is calculated as 0.3, and the similarity between the perfume attribute vector for perfume 3 and the subject attribute vector is calculated as 0.3. The similarity between subject attribute vectors can be calculated as 0.4. The preferred perfume prediction unit 140 checks the similarity value for each of at least one perfume attribute vector, and among these, L perfume attribute vectors whose similarity value is greater than or has a relatively greater similarity value than a predefined reference value. By detecting (s) and determining the perfume corresponding to the detected perfume attribute vector as the preferred perfume for the subject, a prediction result for the preferred perfume can be obtained. Here, L is a natural number greater than 1 and can be defined by the user or administrator. The prediction result for the preferred perfume may be transmitted to the storage unit 103 or the output unit 109. If necessary, the prediction result for preferred perfume may be obtained by combining multiple preferred perfume(s) in the form of a list.

FIG. 12 is a diagram illustrating an example of a judgment process for suitability for dry skin, oily skin, and sensitive skin. In FIG. 12, the solid line means [yes], and the dotted line means [no].

In one embodiment, the preferred perfume prediction unit 140 may determine the preferred perfume for the subject by further using the subject's skin type. That is, as shown in FIG. 11, after or before the similarity value for each perfume is determined, the preferred perfume prediction unit 140 determines whether each perfume is suitable for dry skin, oily skin, and/or sensitive skin, and , prediction results for preferred perfumes can be obtained depending on the suitability of each perfume for each skin type. For example, if a specific perfume is unsuitable for the subject's skin type and/or requires caution, the preferred perfume prediction unit 140 excludes the perfume and obtains a prediction result for the preferred perfume using the remaining perfumes. can do.

According to one embodiment, the preferred perfume prediction unit 140 may determine whether the perfume is suitable using a predetermined algorithm based on the subject's skin type and the characteristics of the perfume, as shown in FIG. 12.

Specifically, first, the preferred perfume prediction unit 140 determines whether the perfume contains irritating chemical ingredients if the subject's skin type is a sensitive skin type, and if the perfume contains irritating chemical ingredients, the perfume is It is judged to be inappropriate. Conversely, if it does not contain irritating chemical ingredients, the preferred perfume prediction unit 140 determines whether it is recommended to spray on parts other than the skin when using the perfume, and if so, determines that the perfume requires caution. Otherwise, The preferred perfume prediction unit 140 may determine that the corresponding perfume is suitable.

If the subject's skin type is not a sensitive skin type but an oily skin type, at least determine whether the perfume uses dark notes. If it uses dark notes, judge it as unsuitable. If it does not use dark notes, determine whether it is suitable for skin soothing. Further determine whether notes of effect are included. If the perfume contains notes having a skin-soothing effect, the preferred perfume prediction unit 140 determines that the perfume is suitable. If the perfume does not contain notes that have a skin-soothing effect, it is determined whether the perfume contains chemicals that are harmful to oily skin. If the perfume contains chemicals that are harmful to oily skin, the perfume is judged to require caution; otherwise, it is judged suitable.

If the subject does not have a sensitive or oily skin type, it is determined whether the perfume contains strong and long-lasting notes. If the perfume includes notes that are strong and have a long-lasting lingering scent, the preferred perfume prediction unit 140 may determine that the perfume is suitable. Conversely, if the perfume is strong and does not contain long-lasting notes, it is additionally determined whether the perfume has moisturizing ingredients, and if the perfume has moisturizing ingredients, it is judged to be suitable. If the perfume in question does not contain strong, long-lasting notes and does not contain moisturizing ingredients, it is judged whether the alcohol content is more than 90%. If the alcohol content is less than 90%, the preferred perfume prediction unit 140 determines that the perfume is suitable for the subject, and if the alcohol content is more than 90%, the perfume is judged to be unsuitable.

The perfume information collection unit 150 may collect at least one perfume information 105. In this case, the perfume information collection unit 150 controls the communication module to connect to a wired or wireless communication network, access one or more perfume-related websites (perfume shopping mall site, social network service site, or community site, etc.), and access Perfume information 105 may be obtained by crawling a website and extracting and collecting information related to at least one perfume from the crawling results. In this case, the perfume-related website may be pre-designated by an administrator or user, or may be determined based on the judgment of the perfume information collection unit 150. For example, the perfume information collection unit 150 may include at least one keyword (e.g., words such as [perfume], [perfume], [eau de toilette], or [shower cologne], a known perfume brand or name, etc. ), a web search may be performed to obtain search results, and perfume-related websites may be detected based on the obtained search results.

The above-described preferred perfume prediction device 100 may be implemented using a device specifically designed to perform a series of processing for predicting preferred perfume, or may be implemented by using at least one information processing device alone or in combination. It could be. Here, information processing devices include desktop computers, laptop computers, server hardware devices, smart phones, tablet PCs, smart watches, navigation devices, portable game consoles, digital televisions, set-top boxes, digital media player devices, media streaming devices, and sound reproduction devices. Devices (artificial intelligence speakers, etc.), manned or unmanned moving vehicles (passenger cars, vehicles such as buses and two-wheelers, mobile robots, wireless model vehicles, robot vacuum cleaners, etc.), manned or unmanned flying vehicles (aircraft, helicopters, drones, model airplanes, etc.) Model helicopters, etc.), household/industrial/military robots, or industrial/military machines, etc.

Hereinafter, an embodiment of a preferred perfume prediction method will be described with reference to FIG. 13.

Figure 13 is a flowchart of an embodiment of a preferred perfume prediction method.

As shown in FIG. 13, perfume information may first be collected (200). The collected perfume information can be stored in the form of a perfume database. Perfume information may be obtained through direct input by a user or administrator, or may be obtained from a server device accessible through a wired or wireless network through methods such as crawling.

Before, after, or simultaneously with the collection of perfume information, information about the subject (i.e., subject information) may be input from the user (202). The subject may be the same as the user or may be different. Target information may include, for example, information about the target individual (target's age, gender, or taste, etc.) or information about the perfume they wish to search for, and may be entered in the form of words, phrases, or sentences. there is.

When target information is input, at least one target keyword may be extracted from the target information (204). Extraction of target keywords can be performed by performing morphological analysis on a given word, phrase, or sentence and, if necessary, removing stop words from the analysis results. Extraction of target keywords can also be performed using, for example, a predetermined learning model.

Once the target keyword is extracted, it may be determined whether the target keyword is a valid attribute keyword (206). Determination of whether the target keyword is a valid attribute keyword may be performed using valid attribute information. Effective attribute information may be previously constructed using at least one word, phrase, or sentence that can be used as a key attribute in predicting preferred perfume. If a valid attribute keyword exists among the target keywords, a related keyword corresponding to the valid attribute keyword is further determined. Determination of related keywords may be performed using a predefined preference matching dictionary.

A target attribute vector may be created using the target keyword and related keywords (208). Specifically, first, the target keyword(s) and the related keyword(s) corresponding to the effective attribute keywords are combined, and as a result, at least one set of keywords is generated. The frequency of each keyword belonging to the keyword set is sequentially measured, and keywords for the attribute vector are extracted from one or more keywords (target keyword and related keyword) according to the frequency of each keyword. In this case, depending on the embodiment, N keywords among one or more keywords may be extracted as keywords for the audience attribute vector according to the order of frequency, or keywords with a frequency greater than the standard frequency may be extracted as keywords for the audience attribute vector. It can also be extracted as a keyword. Keywords for the extracted subject attribute vector are combined, and as a result, a subject attribute vector is generated.

Meanwhile, at least one perfume keyword may be extracted from the perfume information before or after at least one of the target information acquisition or target attribute vector generation processes 202 to 208, or simultaneously with at least one of these processes 202 to 208 ( 210). Extraction of at least one perfume keyword may be performed by performing morphological analysis on each of at least one sentence included in the perfume information and, depending on the embodiment, removing stop words from the morphological analysis results.

Once the perfume keyword is extracted, the importance within the document for a specific word regarding a specific perfume can be calculated (212). Calculation of the importance within a document for a word may, for example, count the word appearance frequency (TF) of at least one specific keyword from at least one document for the perfume in question, and calculate the number of documents in which the keyword appears, i.e. the number of occurrences of the word. This can be done by counting the number of documents (DF) and then dividing the frequency of word occurrence by the number of documents in which the word appears.

Subsequently, a perfume attribute vector can be created using the importance within the document for the word (214). Generation of the perfume attribute vector may be performed by selecting at least one core keyword from at least one perfume keyword, based on the importance of the word in the document, and arranging the at least one core keyword as predefined. Selection of at least one core keyword may be performed by extracting M keywords with greater word importance within the document, or may be performed by extracting keywords with greater word importance within the document than a predefined reference value. Depending on the embodiment, the perfume attribute vector may be created by combining additional extracted keywords. Specifically, the perfume attribute vector counts the frequency of specific core keywords in all or some documents, and keywords whose frequency is relatively large compared to other specific core keywords or whose frequency exceeds a predetermined standard value are additionally extracted as keywords. It may be generated by extracting and obtaining core keywords and arranging the additional extracted keywords obtained according to predetermined standards.

Once the subject attribute vector and the perfume attribute vector are created, prediction of the preferred perfume can be performed based on them (216). Prediction of preferred perfume can also be performed through similarity analysis of the subject attribute vector and perfume attribute vector. Here, the similarity between the subject attribute vector and the perfume attribute vector may be calculated using, for example, Jacquard similarity. If the calculation result value of the Jaccard similarity between the subject attribute vector and the perfume attribute vector is relatively close to 1, the perfume corresponding to the perfume attribute vector may be determined to be a perfume that matches the subject (i.e., preferred perfume). Conversely, if the calculation result value of the Jaccard similarity between the subject attribute vector and the perfume attribute vector is relatively close to 0, the perfume corresponding to the perfume attribute vector may be determined to be a perfume that does not match the subject. If necessary, one-hot encoding may first be applied to the subject attribute vector and the perfume attribute vector.

Depending on the embodiment, the subject's skin type may be further used to determine the preferred perfume. Before or after determining the similarity value, it is determined whether each perfume is suitable for dry skin, oily skin, and/or sensitive skin, and depending on the suitability for each skin type, specific perfumes are excluded and not determined as a preferred perfume. .

Specifically, if the subject's skin type is a sensitive skin type, it is determined whether the perfume contains irritating chemical ingredients, and if the perfume contains irritating chemical ingredients, the perfume is judged to be unsuitable. Conversely, if it does not contain irritating chemical ingredients and requires spraying on areas other than the skin when using the perfume, the perfume is considered to require caution. If the subject's skin type is oily and uses dark notes, the perfume is judged to be unsuitable. If no strong notes are being used, it can be further determined whether the perfume contains notes with a skin-soothing effect, and if the perfume contains notes with a skin-soothing effect, the perfume is judged suitable, and whether the perfume has a soothing effect. If it contains chemicals that are harmful to oily skin without including notes, caution is considered necessary. Otherwise, it is considered suitable. Additionally, if necessary, the perfume may be deemed suitable if it contains strong and long-lasting notes, has moisturizing ingredients, and/or has an alcohol content of 90% or less. If the alcohol content exceeds 90%, the perfume may be deemed unsuitable.

The predicted preferred perfume may be stored in the storage unit, output visually or auditorily through the output unit, or transmitted to another external terminal device (218).

The preferred perfume prediction method according to the above-described embodiment may be implemented in the form of a program that can be driven by a computer device. A program may include instructions, libraries, data files, and/or data structures, etc., singly or in combination, and may be designed and produced using machine code or high-level language code. The program may be specially designed to implement the above-described method, or may be implemented using various functions or definitions that are known and available to those skilled in the art in the computer software field. Here, the computer device may be implemented by including a processor or memory that enables the function of the program, and may further include a communication device as needed. A program for implementing the above-described preference perfume prediction method may be recorded on a recording medium readable by a device such as a computer. Recording media that can be read by a computer include, for example, semiconductor storage media such as ROM, RAM, SD cards, or flash memory (eg, solid state drives (SSD), etc.), or magnetic disk storage such as hard disks or floppy disks. At least one medium capable of temporarily or non-temporarily storing one or more programs that are executed upon call from a device such as a computer, such as an optical recording medium such as a compact disk or DVD, or a magneto-optical recording medium such as a floptical disk. It may include any type of physical storage medium.

The above-described preferred perfume prediction device and preferred perfume prediction method not only predict perfume, but also recommend specific products (e.g., flowers, literary works, etc.) that match the user's taste by analyzing text consisting of poetic or abstract words. It may be applied in the same or slightly modified form to a device or method.

Although several embodiments of the preferred perfume prediction device and the preferred perfume prediction method have been described above, the preferred perfume prediction device or the preferred perfume prediction method is not limited to the above-described embodiments. Various other devices or methods that can be implemented by those skilled in the art by modifying and modifying the above-described embodiments based on the above-described embodiments may also be an embodiment of the above-described preference perfume prediction device or preference perfume prediction method. For example, the described method(s) may be performed in an order other than as described, and/or component(s) of the described system, structure, device, circuit, etc. may be combined, connected, or otherwise in a form other than as described. Even if combined or replaced or substituted by other components or equivalents, it can be an embodiment of the preference perfume prediction device and/or preference perfume prediction method described above.

100: Preferred perfume prediction device 101: Input unit
103: Storage unit 104: Subject information
105: Perfume information 106: Effective attribute information
110: Processor 120: Subject analysis unit
121: Keyword extraction unit 123: Effective attribute determination unit
125: Related keyword acquisition unit 127: Vectorization unit
130: Perfume analysis unit 131: Keyword extraction unit
133: Vectorization unit 140: Preference perfume prediction unit
150: Perfume information collection department

Claims (15)

An input unit that obtains target information and perfume information; and
Extract an audience keyword from the audience information, generate an audience attribute vector using the audience keyword, extract a perfume keyword from the perfume information, generate a perfume attribute vector using the perfume keyword, and generate a perfume attribute vector using the perfume keyword. A preferred perfume prediction device comprising a processor that determines a preferred perfume using similarity between the perfume attribute vectors. According to paragraph 1,
At least one of the subject information and the perfume information includes at least one sentence,
The processor performs morphological analysis on at least one of the target information and the perfume information, and removes stop words from the morpheme analysis result to obtain at least one of the target keyword and the perfume keyword. According to paragraph 1,
When the target keyword is extracted, the processor detects a valid attribute keyword corresponding to a valid attribute among the target keywords, obtains a related keyword corresponding to the valid attribute keyword, and based on the target keyword and the related keyword A preference perfume prediction device that generates the target attribute vector. According to paragraph 1,
When the perfume keyword is extracted, the processor counts the frequency of appearance of the perfume keyword and the number of documents in which the perfume keyword appears in at least one document of the perfume information, and determines the frequency of appearance of the perfume keyword in the document. A preference prediction device that determines the importance of words in a document by dividing their count. According to clause 4,
The processor is configured to obtain at least one core keyword from at least one perfume keyword based on word importance in the document, and generate the perfume attribute vector using the at least one core keyword. According to paragraph 1,
The processor obtains a similarity value by calculating Jaccard similarity between the subject attribute vector and the perfume attribute vector, and corresponds to a perfume attribute vector in which the similarity value is greater than a predefined reference value or the similarity value is relatively larger. A preference perfume prediction device that determines a perfume as a preferred perfume. According to paragraph 1,
The processor further determines a preferred perfume for the subject using the subject's skin type, wherein the subject's skin type includes sensitive or oily. In clause 7,
The processor determines whether the perfume contains irritating chemical ingredients, whether a dark note is used, whether a note with a skin-soothing effect is included, whether a substance harmful to oily skin is included, whether a strong and long-lasting note is included, whether a moisturizing ingredient is included, and A preferred perfume prediction device that determines a preferred perfume using at least one of the alcohol contents. Obtaining target information and perfume information;
extracting target keywords from the target information;
generating a target attribute vector using the target keyword;
extracting perfume keywords from the perfume information;
generating a perfume attribute vector using the perfume keyword; and
A preferred perfume prediction method comprising: determining a preferred perfume using similarity between the subject attribute vector and the perfume attribute vector. According to clause 9,
At least one of the subject information and the perfume information includes at least one sentence,
The step of extracting target keywords from the target information is,
performing morphological analysis on the subject information; and
Preferred perfume prediction method comprising: removing stop words from the morpheme analysis result to obtain the target keyword. According to clause 9,
The step of generating a target attribute vector using the target keyword is,
When the target keyword is extracted, detecting a valid attribute keyword corresponding to a valid attribute among the target keywords;
Obtaining a related keyword corresponding to the valid attribute keyword; and
A preferred perfume prediction method comprising: generating the target attribute vector based on the target keyword and the related keyword. According to clause 9,
The step of generating a perfume attribute vector using the perfume keyword is,
When the perfume keyword is extracted, counting the frequency of appearance of the perfume keyword within at least one document of the perfume information;
Counting the number of documents in which the perfume keyword appears; and
A preference perfume prediction method comprising: determining word importance within a document by dividing the number of documents by the frequency of appearance of the perfume keyword. According to clause 12,
The step of generating a perfume attribute vector using the perfume keyword is,
Obtaining at least one key keyword from at least one perfume keyword based on word importance in the document; and
A preferred perfume prediction method further comprising: generating the perfume attribute vector using the at least one key keyword. According to clause 8,
The step of determining a preferred perfume using the similarity between the subject attribute vector and the perfume attribute vector,
Obtaining a similarity value by calculating Jacquard similarity between the subject attribute vector and the perfume attribute vector; and
A preferred perfume prediction method comprising: determining a perfume corresponding to a perfume attribute vector whose similarity value is greater than a predefined reference value or whose similarity value is relatively greater as a preferred perfume. According to clause 8,
The step of determining a preferred perfume using the similarity between the subject attribute vector and the perfume attribute vector,
Predicting a preferred perfume for the subject using similarity between the subject attribute vector and the perfume attribute vector and the subject's skin type, wherein the subject's skin type includes sensitive or oily. method.