KR20210031653A - The Automatic Recommendation System and Method of the Fashion Coordination - Google Patents

Abstract

The present invention relates to a system and a method for automatically recommending fashion coordination, and more particularly, to a system and a method, which provide fashion coordinations by automatically combining fashion items suitable for a part of fashion items input by a user using different types of the fashion items and vector characteristics. The method of the present invention comprises the steps of: inputting an inquiry item of a user for a fashion in a business operator item DB, and extracting a type and a vector of the inquiry item from the business operator item DB to transmit the same to a reference DB; retrieving and extracting similar items having a vector having a high degree of similarity with the vector of the inquiry item from the reference DB, and then retrieving and extracting reference coordinations including the similar items; extracting reference items included in the reference coordinations from the reference DB, calculating the type and vector for each of the reference items, and transmitting the same to the business operator item DB; retrieving and extracting recommended items having a vector having a high degree of similarity with a vector of the reference items from the business operator item DB; and configuring recommended coordinations by combining the inquiry item and the recommended items in the business operator item DB, and aligning the same to be transmitted to a user. Accordingly, purchase power can be improved while increasing user satisfaction.

Description

The Automatic Recommendation System and Method of the Fashion Coordination}

The present invention relates to a system and method for automatically recommending fashion coordination, and a system and method for providing fashion coordinations by automatically combining fashion items suitable for some fashion items input by a user using a type of fashion item and a vector feature. It is about.

Due to the recent development of internet technology and smart devices, various products are being actively purchased online. In particular, customers who want to purchase fashion clothing online will go through the process of purchasing fashion clothing by accessing a site that sells fashion clothing and browsing the clothing provided by the site, selecting clothing that suits their taste, and making payment. It goes through.

However, since a business operator operating a clothing selling site updates at least hundreds of new clothing items on the site every month in addition to existing clothing items, the range of clothing choices from the customer's point of view expands and increases. There are also difficulties in choosing one.

Therefore, in order to facilitate the customer's convenience in selecting fashion clothing, recently clothing sales sites, etc., provide a service that recommends clothing for customers.

As a method of such a conventional clothing recommendation service, a user-based recommendation method and an item-based recommendation method are used. In the method and the e-commerce recommendation analysis system and recommendation analysis method of Korean Patent Publication No. 10-2014-111225, a user-based recommendation method and an item-based recommendation method are known, respectively.

i) The above user-based recommendation method can be said to be a method of recommending clothing selected by other customers with similar physical conditions, preferences, and tastes to the customer. ii) The above item-based recommendation method is applicable in the past. This is a method of recommending clothes that have similar physical conditions, preferred styles, and tastes to those selected on the site.

However, such conventional methods of recommending clothing have a disadvantage that objectivity is very insufficient because the measurement standard of the similarity of clothing is not constant, and the physical condition of the customer who purchases the product individually for each product, the style and taste of the product, and the product Since it is possible to apply the recommendation method only when additional information such as the number of clicks is provided, there is a problem that the management demand of clothing sellers increases.

On the other hand, recently, there is also an increasing number of sites that provide virtual experience services that allow users to match or experience various fashion clothes in advance. These virtual experience services allow users to select a number of styles one by one to match their size or preference. You can see if it's right, but because you choose too many styles one by one, you can't solve the problem that it takes a lot of time or effort to find product information that suits you.

In addition, as the number of shopping mall sites such as this is increasing and product information that can be compared is increasing, it is increasingly difficult for users to find product information that is more suitable for them. : A method of automatically recommending other items (eg, bottoms, shoes) suitable for a top) is not currently being done.

That is, conventional technologies have a disadvantage in that they cannot accurately satisfy customer interest and satisfaction by relying on fragmentary data such as HTML data, search keywords, click view status, and past purchase history of a corresponding customer.

Therefore, fashion items that minimize the effort and time of customers purchasing fashion items and increase purchase desire by automatically sorting and displaying recommended items to complete fashion coordination for fashion items that the customer likes. Recommendation method is required.

The present invention was conceived to solve the problems of the prior art described above, and by automatically combining other fashion items suitable for some fashion items input by the user and presenting fashion coordination, the user can easily evaluate and purchase fashion items. It is an object of the present invention to provide an automatic recommendation system for fashion coordination and an automatic recommendation method.

In addition, the present invention promptly arranges similar items, reference items, and recommended items using the type and vector characteristics of fashion items to present a number of completed fashion coordinates, thereby presenting fashion coordinates that are satisfactory to the user in a short time. Another object is to provide an automatic fashion coordination recommendation system and an automatic recommendation method capable of improving purchasing power while increasing user satisfaction.

The technical problems to be achieved by the present invention are not limited to the technical problems mentioned above, and other technical problems that are not mentioned will be clearly understood by those of ordinary skill in the art from the description of the present invention. .

According to an aspect of the present invention devised to solve the problems of the prior art described above, (A) a query item of a user regarding fashion is input into a business item DB, and the type of the query item in the business item DB (part) And extracting a vector and transmitting it to a reference DB. (B) searching and extracting similar items having a vector having a high similarity to the vector of the query item in the reference DB, and then searching and extracting reference coordinations including the similar items; (C) extracting reference items included in the reference coordinations from the reference DB, calculating a type and vector for each of the reference items, and transmitting them to a service provider item DB; (D) searching and extracting recommended items having a vector having a high similarity to the vector of the reference items in the business item DB; It provides an automatic recommendation method for fashion coordination including (E) combining the query items and recommended items in the business item DB to form recommended coordinations, sorting them, and transmitting them to the user.

In the present invention, it is preferable that the step (A) extracts the type and vector of the query item from the query item vector calculation module of the operator item DB and transmits it to a reference DB.

In the present invention, step (B) includes the steps of (B1) searching and extracting similar items having a vector having a high similarity to the vector of the query item in the similar item search module of the reference DB, and (B2) the reference DB It is preferable that the reference coordination search module is formed by searching and extracting reference coordinations including the similar items.

In the present invention, step (C) includes: (C1) separating and extracting reference items included in reference codes through a reference item extraction module of the reference DB, and (C2) a reference item vector calculation module of the reference DB. It is preferable to form a step of calculating a type and a vector for each of the reference items through and transmitting to the operator item DB.

In the present invention, in step (D), it is preferable to search and extract recommended items having a vector having a high similarity to the vectors of reference items in the recommended item search module of the business item DB.

In the present invention, the step (E) comprises the steps of (E1) combining a query item and recommended items in a recommended coordination configuration module of the business item DB to form recommended coordinations, and (E2) in a recommended coordination arrangement module of the business item DB. It is preferable that the recommended coordinates are arranged and displayed to the user in the sorted order.

In the present invention, in step (C), it is preferable to remove an item having the same type as the query item from among the reference coordinates, and extract an item having a different type from the query item as a reference item.

In the present invention, in step (C), it is preferable to search and extract reference coordinates in the order of relatively high quality scores of the previously inputted coordinates through the reference coordinate quality evaluation module of the reference DB.

In the present invention, the step (E) includes multiplying two or more factors among a quality score of a reference coordination, a similarity value of a similar item, or a similarity value of a recommended item, in descending order, based on a result obtained by multiplying or calculating an average value. It is desirable to sort and send to the user.

In the present invention, in step (B), it is preferable that the similar item search module of the reference DB searches and extracts 5 to 10 similar items having a vector having a high similarity to the vector of the query item.

In the present invention, in step (D), it is preferable to search and extract 5 to 10 recommended items having vectors having a high similarity to the vectors of reference items in the recommended item search module of the business item DB.

According to another aspect of the present invention devised to solve the above-described problems of the prior art, a query item vector calculation module for extracting a type (part) and a vector of an input user's query item and transmitting it to a reference DB; A recommended item search module for searching and extracting recommended items having a vector having a high similarity to a vector of reference items input from the reference DB; A recommendation coordination configuration module that combines the user's query item and recommendation items to form recommendation coordinations; And a recommended coordination sorting module for sorting the recommended coordinations and transmitting them to the user in the sorted order; searching and extracting similar items having a vector having the highest similarity with the vector of the query item and a business item DB including. A similar item search module; A reference coordination search module for searching and extracting reference coordinations including the similar items; A reference item extraction module for extracting reference items included in the reference coordinates, respectively; And a reference item vector calculation module that calculates the type and vector for each of the reference items and then transmits it to a business item DB.

In the present invention, it is preferable that the reference DB further includes a reference coordination quality evaluation module that searches and extracts reference coordinations in consideration of the quality scores of the coordinations.

In the present invention, the recommended coordination sorting module, based on a result obtained by multiplying or calculating an average value of any two or more of a quality score of a reference coordination, a similarity value of a similar item, or a similarity value of a recommended item, in descending order. It is desirable to sort and send to the user.

In the present invention, the reference coordination is preferably a set of only coordinations in which a plurality of coordinations have been presented to the consumer in advance, and then the consumers have given a score of more than a predetermined criterion.

In the present invention, the type (part) of the item is preferably classified into tops, bottoms, coats, shoes, bags, and other accessories.

In the present invention, it is preferable that the query item is input through a user terminal or through a display panel connected to a business item DB.

In the present invention, it is preferable that the similar item has the same part as the query item.

* In the present invention, the coordination is preferably a fashion set including at least a top and bottom, or a fashion set including any one or more of shoes, coats, bags, and other accessories on the top and bottom.

According to the automatic fashion coordination recommendation system and automatic recommendation method of the present invention, by automatically combining other fashion items suitable for some fashion items input by the user and presenting fashion coordination, the user can easily evaluate and purchase fashion items. It works.

In addition, according to the present invention, a number of completed fashion coordinates are presented by quickly arranging similar items, reference items, and recommended items using the type and vector characteristics of fashion items, thereby providing fashion coordinates that are satisfactory to the user in a short time. By presenting it, there is an effect of improving the purchasing power while increasing the user's satisfaction.

1 is a configuration diagram of an automatic fashion coordination recommendation system according to an embodiment of the present invention.
2 is an exemplary view showing a state in which fashion items are mapped to a vector space according to an embodiment of the present invention.
3 is a flowchart of a method for automatically recommending fashion coordination according to an embodiment of the present invention.
4 is a flow chart of various data for each step of the system for automatically recommending fashion coordination according to an embodiment of the present invention.
5 is an exemplary view showing a state in which recommended coordination is displayed after inputting a query item according to an embodiment of the present invention.
6 is an exemplary view showing a state in which a recommended coordination in which a query item and a recommended item are combined according to an embodiment of the present invention is displayed.
7 is an exemplary view showing a state in which query items and recommended items are combined and arranged according to an embodiment of the present invention.
8 is an exemplary view showing a state in which recommended coordinates are displayed on a user terminal according to an embodiment of the present invention.
9 is an exemplary view showing a state in which a query item input and recommendation coordinates are displayed according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms or words used in the present specification and claims should not be construed as limited to their usual or dictionary meanings, and the inventors appropriately explain the concept of terms in order to explain their own invention in the best way. Based on the principle that it can be defined, it should be interpreted as a meaning and concept consistent with the technical idea of the present invention. Accordingly, the embodiments described in the present specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical spirit of the present invention. It should be understood that there may be equivalents and variations.

In order to solve the problems of the prior art, the present invention applies a deep learning technology to fashion items (query items) that the user likes or is interested in, and the types of fashion items and mapped vector features After sequentially extracting similar items, reference items, and recommended items by using, a number of recommended coordinations (hereinafter referred to as'coordination' or'cody') are quickly and automatically combined by combining the recommended items with the user's query item. We present an automatic recommendation system and automatic recommendation method for the proposed fashion coordination.

1 is a block diagram of an automatic fashion coordination recommendation system according to an embodiment of the present invention.

The present invention is configured to include a business item DB (100) and a reference DB (200).

The operator item DB 100 includes a query item vector calculation module 110 that extracts a type (part) and a vector of an input user's query item and transmits it to the reference DB 200, and the reference DB ( A recommended item search module 120 that searches for and extracts recommended items having a vector with a high similarity to the vector of reference items input from 200), and a recommended coordination configuration that combines the user's query items and recommended items to form recommended coordinates. It may be configured to include a module 130 and a recommended coordinate alignment module 140 that sorts the recommended coordinates and transmits the recommended coordinates to the user in the sorted order.

Meanwhile, the reference DB 200 searches for and extracts similar items having a vector having the highest similarity with the vector of the query item, and a reference coordination including the similar items. The reference coordination search module 220 to be extracted, the reference item extraction module 230 for extracting reference items included in the reference coordinations respectively, and a business item after calculating the type and vector for each of the reference items It may be configured to include a reference item vector calculation module 240 transmitted to the DB.

In addition, the reference DB 200 may further include a reference coordination quality evaluation module 250 for searching and extracting reference coordinations in consideration of the quality scores of the coordinations.

The user's query item 410 may be an item that the user is interested in or has a good feeling while searching for fashion items, and among a plurality of fashion items on the user terminal 300, a query item to be inquired to the business item DB 100 ( 410) or input through a display panel (not shown) connected to the operator item DB.

In the present invention, a query item 410, a similar item 420, a reference item 440, and a recommended item 450 can be extracted and used. Part can be classified into tops, bottoms, coats, shoes, bags, and other accessories.

However, in the present invention, it is not limited to the classification of the above item types, and it is also possible to be classified into the types of men's and women's items according to the needs of the invention.i) For men, shirts, t-shirts, suit tops, and bottoms , Cardigan, coat, shoes, bag, waistband, tie, tie pin, watch, hat, etc. , Shoes, bags, scarves, shawls, watches, necklaces, bracelets, hats, etc. The vector (vector) extraction and use of the item will be described later in detail.

The similar item 420 is preferably the same as the query item 410 and the same part of the item. For example, if the query item 410 relates to a shirt item among the tops, the similar item 420 is also Can be extracted as a shirt item.

By the reference DB 200, the similar item 420 can extract the similar items 420 having the highest similarity by comparing the vector with the vector of the query item 410, and It is possible to extract 5 to 10 similar items in order, but is not necessarily limited thereto, and similar items may be extracted within a range of 5 to 100 depending on the needs of the invention.

In the present invention, the cody may be defined as a fashion set including at least a top and bottom, or a fashion set including any one or more of shoes, coats, bags, and other accessories in the top and bottom.

That is, the coordination (cody) should at least include a top and bottom, and furthermore, a fashion set coordination can be formed by adding any one or more of shoes, coats, bags, and other accessories to the top and bottom according to the customer's request, etc. There will be.

The reference coordination 430 refers to a coordination including the similar item 420 extracted by the reference DB 200. If the extracted similar item 420 is a T-shirt, as a fashion set including the corresponding T-shirt, it may be referred to as a fashion set further including bottoms, shoes, and the like.

The reference coordinations 430 are stored in the reference DB 200 in advance. After presenting a plurality of coordinations to a consumer in advance, a set of only coordinations to which the consumers have given a score equal to or higher than a predetermined criterion may be collected and stored. For example, after presenting hundreds or thousands of coordinations to a consumer who participated in a survey, if the satisfaction score of the consumer exceeds 50 points, it may be adopted as the reference coordination 430 and stored.

The consumer's satisfaction score may be metadata collected at the same time when data is collected for the reference coordinator 430, and may be a score assigned by a user of various fashion sites or a value generated based on the number of clicks on the favourite.

The reference item 440 is an item belonging to the reference coordination 430 and refers to an item having a type different from that of the query item 410. That is, if the query item 410 is a top, the reference item 440 cannot be a top, and only bottoms, shoes, and other accessories will be extracted as a reference item.

The recommended items 450 are items having the same type as the reference item 440 and having a high vector similarity, and among items stored in the business item DB 100, about 5 to 100 items having a high vector similarity to the reference item 440 Among the items belonging to the range of may be extracted items of the range requested by the user.

Since the recommended item 450 is an item having the same type as the reference item 440, an item of a type different from the type of the query item 410 will be extracted.

The recommended coordination 460 is a combination of the query item 410 and the recommended item 450, and since both are items of different types, they can be directly combined into a fashion set.

The recommended coordination arrangement module 140 performs a function of arranging a plurality of recommended coordinations 460 and displaying it to the user. In this case, the criteria for alignment are the quality score of the reference coordination, the similarity value of the similar item, or the recommended item. According to the result obtained by multiplying any two or more of the similarity values of or calculating the average value, it is sorted in descending order and transmitted to the user.

That is, the sorting criteria of the recommended coordination 460 are [(Quality score of reference coordination) X (Similarity value of similar item)], [(Quality score of reference coordination) X (Similarity value of recommended item)], [ Based on the result of either (similarity value of similar item) X (similarity value of recommended item)] or [(Quality score of reference coordination) X (similarity value of similar item) X (similarity value of recommended item)] You can sort in descending order.

In addition, according to the necessity of the invention, the sorting criterion of the recommended coordination 460 may be'average value of the quality score of the reference coordination and the similarity value of the similar item' or'average value of the quality score of the reference coordination and the similarity value of the recommended item' or Sort in descending order based on either the result of'the average value of the similarity value of similar items and the similarity value of the recommended item' or'the quality score of the reference coordination, the similarity value of similar items, and the average value of the similarity value of recommended items' can do.

Here, the average value may be an average value of any one of an arithmetic average value, a geometric average value, and a harmonic average value. In some cases, the average value may be replaced with a linear sum calculated by a linear combination.

The'quality score of reference coordination' is an average score given by consumers participating in the survey, the'similarity value of similar items' means a vector similarity value with a query item, and the'recommended item similarity value' is a reference It means the value of the vector similarity to the item.

However, the sorting criteria of the recommended coordinates 460 are not limited to the above example, and further combinations of query items, similar items, reference items and recommended item individual product attributes, customer satisfaction, or vector features according to the need of the invention. So it could be used as a standard for sorting.

Here, the satisfaction level of the consumer may be metadata that is collected together when information on each item is collected, and is based on scores given to individual items by users of various fashion sites, the number of clicks on the like mark, the number of comments, and page views. It may be a value generated by

The business item DB 100 and the reference DB 200 may simultaneously perform a function of a database storing data therein and a function of a server controlling data flow including various functional modules.

2 is an exemplary view showing a state in which fashion items are mapped to a vector space according to an embodiment of the present invention.

The business item DB 100 and the reference DB 200 of the present invention may store images and information of various fashion items, and may collect and store images and information of fashion items from domestic and international fashion shopping mall sites.

Meanwhile, the business item DB 100 and the reference DB 200 may map the image data of the fashion item 510 to a specific location 530 on the preset vector space 520, from which the type of the fashion item (Category vector) and vector information can be obtained.

And, looking at an example of a method of extracting an item with similar vector characteristics from a specific item, first, by mapping the image data of a specific item to a specific location in a preset vector space, a basic vector corresponding to the image data is obtained, After obtaining a similar vector using the obtained basic vector and the category vector of a specific item, image data of a fashion item different from the specific item may be obtained based on the obtained basic vector.

For reference, the'category vector' may be a vector determined based on vectors in a predetermined area of a vector space corresponding to a category of a fashion item. That is, the category vector may be a vector corresponding to the average of vectors in a predetermined area of a vector space corresponding to the category of fashion items. For example, the category vector of the fashion product category'long-sleeved top' is a fashion product called a long-sleeved top. It may be a vector corresponding to an average of vectors on a plurality of long sleeves within a predetermined area of the vector space corresponding to the category.

Meanwhile, in the present invention, the fashion item vector may be a feature vector extracted from an image of each item, and may be extracted by applying a CNN-based image feature extraction method.

In addition, in the present invention, the similarity between various items can be calculated by applying a vector similarity calculation method such as cosine similarity and jaccard similarity.

3 is a flowchart of a method for automatically recommending fashion coordination according to an embodiment of the present invention, and FIG. 4 is a flow chart of various data for each step of the automatic fashion coordination system according to an embodiment of the present invention.

In the present invention, (A) a user's query item 410 related to fashion is input to the operator item DB 100, and the type (part) and vector of the query item 410 in the operator item DB 100 Extracting and transmitting it to the reference DB 200; (B) After searching and extracting similar items 420 having a vector having a high similarity to the vector of the query item 410 in the reference DB 200, a reference coordinate 430 including the similar items 420 ) Searching and extracting; (C) After extracting the reference items 440 included in the reference coordinates 430 from the reference DB 200, the type and vector are calculated for each of the reference items 440, and the operator item DB ( Transmitting to 100); (D) searching and extracting recommended items 450 having a vector having a high similarity to the vectors of the reference items 440 in the business item DB 100; (E) combining the query item 410 and the recommended item 450 in the business item DB 100 to form the recommended coordination 460, sorting it, and transmitting it to the user; automatic fashion coordination including Provide recommendations.

A detailed look at the step-by-step sequence of the method for automatically recommending fashion coordination according to the present invention is as follows.

First, a user's query item 410 regarding fashion is input to the business item DB 100 (S11). The query item 410 may be selected and input on the user terminal 300 or may be input through a display panel connected to the business item DB 100 or the like. The user may be said to input a query item to receive recommendations for other items matching the query item 410 selected by the user, and eventually, the recommendation coordination 460 by combining the recommended items 450 suitable for the query item 410. ) Is presented to the user.

Subsequently, the query item vector calculation module 110 of the business item DB extracts the type and vector of the query item 410 and transmits it to the reference DB 200 (S12).

Referring to FIG. 4, as an example of the query item vector calculating module 110, a query item vector calculating table v_dict_vdb, which is a type of hash table, is illustrated.

In the present invention, all items have an item identification number (iid), and such an item identification number (iid) may be stored in the business item DB 100 and the reference DB 100, respectively.

The query item vector calculation table v_dict_vdb may perform a function of rapidly searching for and extracting a part and a vector of an item based on a unique number (iid) value of a query item.

Subsequently, the similar item search module 210 of the reference DB searches and extracts similar items 420 having a vector having a high similarity to the vector of the query item 410 (S13).

Referring to FIG. 4, as an example of the similar item search module 210, a similar item search table r_sep_vdb, which is a type of hash table, is shown. The similar item search table r_sep_vdb extracts a unique number (iid) of the similar item 420 having a high vector similarity to the query item 410 and transmits it to the reference coordination search module 220.

The similar item search table (r_sep_vdb) is indexed by separating by type in order to efficiently perform a similarity search based on a vector.For fast vector search, an indexing function such as Locality Sensitive Hashing is performed. I can.

The similar item search module 210 searches for and extracts 5 to 10 similar items 420 having a vector having a high similarity to the vector of the query item 410 according to the need of the invention, and uses the reference coordination search module 220. I will be able to deliver it.

Thereafter, the reference coordination search module 220 performs a step of searching and extracting reference coordinations 430 including similar items 420 (S14).

In the present invention, a set of reference coordinations 430 is previously stored in the reference DB 200, and each reference cody has a unique number (cid), and the unique number (cid) is also a reference DB (200). ) Can be stored.

Referring to FIG. 4, as an example of the reference coordination search module 220, a reference coordination search table r_iid_to_cid, which is a type of hash table, is shown. The reference coordination search table r_iid_to_cid performs a function of extracting the unique number (cid) of the reference coordination 430 including the extracted similar item 420 and transmitting it to the reference item extraction module 230.

Meanwhile, in this case, the reference coordination quality evaluation module 250 may search and extract the reference coordination 430 in the order of a relatively high quality score of the previously input coordination. As described above, the quality score of the coordinates may be an average score evaluated by the consumers after presenting a plurality of coordinates to the consumer.

Subsequently, a step of separating and extracting reference items 440 included in the reference coordinates 430 through the reference item extraction module 230 is performed (S15).

Referring to FIG. 4, as an example of the reference item extraction module 230, a reference item extraction table r_cid_to_iid, which is a type of hash table, is shown.

The reference item extraction table (r_cid_to_iid) extracts the identification number (iid) of the reference item 440 belonging to it based on the identification number (cid) of the extracted reference coordination 430, and the reference item vector calculation module 240 It performs the function of transferring to.

However, the reference item extraction module 230 removes an item having the same type as the query item 410 from among the reference coordinates 430, and selects an item having a different type from the query item 410 as the reference item 440. ). This is to present a recommended coordination by combining different types of items suitable for the query item 410 while leaving the query item 410 that the user is interested in.

Thereafter, the type and vector are calculated for each of the reference items 440 through the reference item vector calculation module 240 and then transmitted to the recommended item search module 120 of the operator item DB (S16). ).

Referring to FIG. 4, as an example of the reference item vector calculation module 240, a reference item vector calculation table r_dict_vdb, which is a type of hash table, is shown.

The reference item vector calculation table r_dict_vdb performs a function of calculating a type and a vector based on the identification number iid of the reference item 440 and transmitting it to the recommended item search module 120.

Then, the recommended item search module 120 of the business item DB 100 searches for and extracts recommended items 450 having a vector having a high similarity to the vectors of the reference items 440 (S17).

Referring to FIG. 4, as an example of the recommended item search module 120, a recommended item search table v_sep_vdb, which is a type of hash table, is shown.

The recommended item search table (v_sep_vdb) extracts the identification numbers (iid) of the recommended items 450 having a relatively high vector similarity to the reference items 440 and transfers them to the recommended coordination configuration module 130. It is done.

The recommended item search table (v_sep_vdb) is also indexed by separating by type in order to efficiently perform similarity search based on a vector, and indexing function is performed in a manner such as Locality Sensitive Hashing for fast vector search. I can.

The recommended item search module 120 may search and extract 5 to 10 recommended items 450 having vectors having a high similarity with the vectors of the reference items 440 and transmit them to the recommended coordination configuration module 130.

Subsequently, the recommendation coordination configuration module 120 combines the query item 410 and the recommendation item 450 to configure a plurality of recommendation coordinations (S18).

Since the business item DB already includes both the unique number (iid) of the query item and the unique number (iid) of newly delivered recommended items, the recommended coordination 460 based on the unique number (iid) of the query item and the recommended items. ) Can be easily created.

Then, the recommended coordination arrangement module 140 performs a step of aligning the recommended coordination 450 according to a predetermined criterion (S19). That is, the recommended coordinates are sorted in descending order and transmitted to the user based on a result obtained by multiplying or calculating an average value of any two or more of the quality score of the reference coordination, the similarity value of the similar item, or the similarity value of the recommended item.

The number of the sorted recommended coordinations may be 5 to 10, but is not necessarily limited thereto, and any number may be adopted as long as it is within the range of 5 to 100 according to the need of the invention.

Finally, a step of displaying the sorted recommended coordinates 450 to the user is performed (S20).

That is, the recommended coordinates 450 can be transmitted from the recommended coordinate arrangement module 140 to the user terminal 300 in sorted order, or displayed to the user through a display screen connected to or attached to the operator item DB 100 It would be possible to do it.

5 is an exemplary view showing a state in which recommended coordination is displayed after inputting a query item according to an embodiment of the present invention.

5 is a compressively showing the automatic recommendation method of fashion coordination of the present invention. When a user selects and inputs a T-shirt as a query item 410, a T-shirt with a high vector similarity to the query item 410 is a similar item ( 420).

In this way, after searching for and extracting the reference coordination 430 including the T-shirt extracted as the similar item 420, among the reference coordinations 430, the similar item 420 having the same type as the query item 410 is removed, Only reference items 440 of different types are extracted, and a recommended item 450 having a high vector similarity is searched and extracted.

That is, the recommended items 450 are limited to bottoms, shoes, etc., which are different from the query item 410, so in the end, the quality item 410, which is a T-shirt, is combined with the recommended item 450, which is a bottom, and the shoes are recommended. 460 is configured and displayed to the user.

5 shows only one recommendation coordination 460, but if the user does not like the recommended coordination 460 displayed, a plurality of recommendation coordinations 460 arranged in descending order are quickly presented and the user can browse. You can do it.

6 is an exemplary view showing a state in which a recommended coordination in which a query item and a recommended item are combined according to an embodiment of the present invention is displayed.

6, when a user inputs a blouse as the query item 410, it can be seen that a plurality of skirts are recommended and presented as a recommended item 450 suitable therefor.

In the present invention, the recommended coordinates 460 may be displayed one by one according to the setting of the user terminal 300, or a plurality of recommended coordinates 460 may be displayed at once.

On the other hand, the result obtained by multiplying or calculating the average value of any two or more of the quality score of the reference coordination, the similarity value of the similar item, or the similarity value of the recommended item by the recommendation coordination sorting module 140 of the business item DB 100 Since the recommended coordinates 460 are arranged in descending order based on and displayed to the user, there is an advantage in that the quality score for each recommended coordinate 460 can be simultaneously displayed as shown in the lower part of FIG. 6.

7 is an exemplary view showing a state in which query items and recommended items are combined and arranged according to an embodiment of the present invention.

7 shows a recommended coordination based on a result obtained by multiplying or calculating an average value of any two or more of a quality score of a reference coordination, a similarity value of a similar item, or a similarity value of a recommended item by the recommended coordination arrangement module 140 460) are arranged in descending order.

At this time, when a plurality of recommended items 450 are combined with the query item 410, [(Quality score of reference coordination) X (Similarity value of similar items)], [(Quality score of reference coordination) for each recommended item ) X (Similarity value of recommended item)], [(Similarity value of similar item) X (Similarity value of recommended item)] or [(Quality score of reference coordination) X (Similarity value of similar item) X (Similarity value of recommended item) Similarity value)] may be averaged to finally calculate a quality score for each recommended coordination.

In addition, as described above, according to the necessity of the invention, the sorting criteria of the recommended coordination 460 may be'average value of the quality score of the reference coordination and the similarity value of the similar item' or'the quality score of the reference coordination and the similarity value of the recommended item. Based on either the result of'the average value of' or'the average value of the similarity value of the similar item and the similarity value of the recommended item' or'the quality score of the reference coordination, the similarity value of the similar item, and the average value of the similarity value of the recommended item' You can also sort by.

FIG. 8 is an exemplary diagram showing a state in which recommendation coordinates are displayed on a user terminal according to an embodiment of the present invention, and FIG. 9 is an exemplary diagram showing a state in which a query item input and recommendation coordinates are displayed according to an embodiment of the present invention. .

That is, in the present invention, the user may access the user item DB 100 and select and input any one of the various items stored in the user item DB 100 as the query item 410, and during an actual Internet search. A recommended coordination 460 may be presented by selecting and inputting a found item or an item taken at a store as the query item 410.

If the business item DB 100 does not have the identification number (iid) of the item found during the actual Internet search or the item taken at the store, the query item is replaced with the item having the closest vector similarity, and then recommended. The coordination 460 may be configured and presented.

As described above, according to the present invention, fashion coordination is presented by automatically combining other fashion items suitable for some fashion items input by the user, but quickly similar items, reference items, and recommended items using the type of fashion item and vector characteristics. By arranging and presenting a number of completed fashion coordinations, there is an advantage of improving purchasing power while increasing user satisfaction by presenting a plurality of recommended coordinations with a quality score guaranteed in a short time.

The present invention has been described above in connection with the specific embodiments of the present invention, but this is only an example and the present invention is not limited thereto. Those of ordinary skill in the technical field to which the present invention pertains may change or modify the described embodiments without departing from the scope of the present invention, and within the scope of the technical idea of the present invention and the claims to be described below. Various modifications and variations are possible.

100: user item DB
110: query item vector calculation module
120: recommended item search module
130: recommended coordination configuration module
140: recommended coordination alignment module
200: reference DB
210: similar item search module
220: reference coordination search module
230: Reference item extraction module
240: reference item vector calculation module
250: reference coordination quality evaluation module
300: user terminal
410: query item
420: Similar Items
430: reference coordination
440: reference item
450: Recommended item
460: recommended coordination
470: Recommended coordination evaluation score
510: fashion item
520: vector space
530: specific position in vector space

Claims (5)

(A) Through the user terminal, the user's query item related to fashion is input into the business item DB, and a reference DB by extracting the part and vector of the query item from the query item vector calculation module of the business item DB Transferring to;
(B) After searching and extracting 5 to 10 similar items having a vector having a high similarity with the vector of the query item in the similar item search module of the reference DB, the similar items are searched by the reference coordination search module of the reference DB. Searching and extracting the included reference coordinates;
(C) After separating and extracting reference items included in the reference coordinates through the reference item extraction module of the reference DB, the type and vector for each of the reference items through the reference item vector calculation module of the reference DB Calculating and transmitting to a business item DB;
(D) searching and extracting 5 to 10 recommended items having vectors having a high similarity to the vectors of the reference items in the recommended item search module of the business item DB; And
(E) The recommendation coordination configuration module of the business item DB combines the query item and the recommended items to compose recommended coordinations, and the recommendation coordination arrangement module of the business item DB arranges the recommended coordinations, sorts them, and sends them to the user. Including;
The (E) step,
In the recommended coordination sorting module of the business item DB, recommended coordinations are arranged in descending order based on a linear sum of the quality score of the reference coordination, the similarity value of the similar item, and the similarity value of the recommended item. Sorted and sent to the user,
The (C) step,
Through the reference coordination quality evaluation module of the reference DB, reference coordinations are searched and extracted in the order of relatively high quality scores of previously input coordinations,
An automatic recommendation method for fashion coordination, characterized in that, among the reference coordinations, an item having the same type as the query item is removed, and an item having a type different from the query item is extracted as a reference item.
In the automatic recommendation system for fashion coordinates that performs the method of automatically recommending fashion coordinates of claim 1,
A query item vector calculation module that extracts a part and a vector of an input user's query item and transmits it to a reference DB;
A recommended item search module for searching and extracting recommended items having a vector having a high similarity to a vector of reference items input from the reference DB;
A recommendation coordination configuration module that combines the user's query item and recommendation items to form recommendation coordinations; And
A business item DB including; a recommended coordination sorting module that sorts the recommended coordinations and transmits the recommended coordination to the user in the sorted order,
A similar item search module for searching and extracting similar items having a vector having the highest similarity with the vector of the query item;
A reference coordination search module for searching and extracting reference coordinations including the similar items;
A reference item extraction module for extracting reference items included in the reference coordinates, respectively;
A reference item vector calculation module that calculates a type and a vector for each of the reference items and transmits it to a business operator item DB; And
A reference coordinate quality evaluation module that searches and extracts reference coordinates in consideration of the quality scores of the coordinates; Having a reference DB including,
The recommended coordination sorting module sorts recommended coordinations in descending order based on a linear sum of a quality score of a reference coordination, a similarity value of a similar item, and a similarity value of a recommended item. Transfer to, and
The reference coordinations are only a set of coordinations in which a number of coordinations have been presented to the consumer in advance, and then the consumers have given a score of more than a predetermined standard
The automatic recommendation system for fashion coordination, wherein the similar item has the same type of item as the query item.
The method of claim 2, wherein the type (part) of the item is,
Automatic recommendation system for fashion coordination, characterized in that it is classified into tops, bottoms, coats, shoes, bags, and other accessories.
The method of claim 2, wherein the query item,
An automatic recommendation system for fashion coordination, characterized in that input through a user terminal or through a display panel connected to a business item DB.
The method of claim 2, wherein the coordination is
An automatic recommendation system for fashion coordination, characterized in that it is a fashion set including at least a top and bottom or a fashion set including one or more of shoes, coats, bags, and other accessories on top and bottom.

Images