KR102064589B1 - System and method for recommending goods - Google Patents

Abstract

The product recommendation server according to an exemplary embodiment includes a communication unit, a review database storing review information about a plurality of products, and a controller electrically connected to the communication unit and a review database, wherein one review information corresponding to one product is At least one review content and at least one rating information, the control unit is a product that includes a keyword received through the communication unit of the plurality of products as a tag, the latest information, frequency of review information corresponding to each of the plurality of products, And recommend products that are within a critical ranking when listing a plurality of products in order of ranking according to the evaluation value calculated using the rating.

Description

Product recommendation server and product recommendation method {SYSTEM AND METHOD FOR RECOMMENDING GOODS}

The present disclosure relates to a product recommendation server and a product recommendation method.

Recently, various products and services have been sold in an electronic commerce method. In this e-commerce scheme, consumers obtain information about products and services to purchase by referring to existing buyers' evaluation of products and services.

However, as the types of products and the information to be referred to for each product increase, consumers may have difficulty in selecting a product that meets their needs.

In order to solve this problem, a product recommendation system is introduced that provides a list of recommended products to consumers to help consumers easily find a product that is highly likely to be purchased.

The present disclosure aims to solve the above and other problems. Another object is to provide a product recommendation server and a product recommendation method for recommending a product meeting a user's needs.

In order to achieve the above or another object, a product recommendation server according to an embodiment includes a communication unit, a review database storing review information about a plurality of products, and a control unit electrically connected to the communication unit and a review database. The review information corresponding to the product includes at least one review content and at least one rating information, and the controller is a product including a keyword received through a communication unit among a plurality of products as a tag, and corresponds to each of the plurality of products. When listing a plurality of products in order of ranking according to the evaluation value calculated using the recency, frequency, and rating of the review information to recommend a product within the threshold ranking.

The controller extracts a word having a frequency greater than or equal to a threshold from at least one review content of one review information as a tag, calculates a weight for the tag according to the degree of use of the tag in the review information about the plurality of products, The tag value corresponding to the tag may be calculated by multiplying the frequency value by a weight.

The control unit may determine the equality of the number of at least one review content of one review information among the distribution of the number of contents of each of the review information about the plurality of products, a time point at which each of the at least one review content of one review information is created, and at least One piece of rating information may be used to calculate an RFS value of one item.

The controller searches for a product having a tag corresponding to a keyword among the plurality of products, calculates an evaluation value by using the RFS value of the searched product and the tag value of the tag corresponding to the keyword of the searched product, and according to the evaluation value, It is possible to determine the ranking of the products of, and to recommend products corresponding to the top ranking.

,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 빈도,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 개수,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 모든 단어들의 개수일 수 있다.The control unit calculates the frequency of each of the words in the at least one review content of one review information using the following equation,

,

Is the frequency of a particular word in at least one review content of one review information,

The number of specific words in at least one review content of one review information,

May be the number of all words in at least one review content of one review information.

,

는 태그에 대한 가중치,

는 복수의 상품에 대한 리뷰 정보들의 전체 개수,

는 복수의 상품에 대한 리뷰 정보들 중 특정 단어를 포함하는 리뷰 정보의 개수일 수 있다.The control unit calculates the weight for the tag using the following equation,

,

Is the weight for the tag,

Is the total number of review information for a plurality of products,

May be the number of review information including a specific word among review information about a plurality of products.

, WF는 복수의 상품에 대한 리뷰 정보들 각각의 컨텐츠의 개수의 분포 중 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠의 개수의 등위에 따라 결정된 가중치, WR은 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠 각각이 작성된 시점에 따라 결정된 가중치, 및 S는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠 각각에 대응하는 평점일 수 있다.The controller calculates an RFS value of one product using the following equation,

WF is a weight determined according to the equality of the number of at least one review content of one review information among the distribution of the number of contents of each of the review information for a plurality of products, and WR is at least one review content of one review information. A weight determined according to a time point at which each is written, and S may be a rating corresponding to each of at least one review content of one review information.

The controller may calculate an evaluation value by multiplying a tag value of a tag corresponding to a keyword of the searched product by a square root of the RFS value of the searched product.

According to an embodiment, a method of recommending a product includes reading review information about a plurality of products from a review database, wherein one review information corresponding to one product includes at least one review content and at least one rating information Receiving a keyword through the communication unit, searching for a product including the keyword as a tag among the plurality of products, and using the latest value, frequency, and rating of the review information corresponding to each of the searched products. And determining products which are within a threshold ranking as recommended products when listing products searched in the order of ranking according to the ranking.

The determining of the recommendation product may include extracting, as a tag, a word having a frequency greater than or equal to a threshold value from at least one review content of one review information, and determining the recommendation product according to the degree of use of the tag in the review information about the plurality of products. The method may include calculating a weight, and calculating a tag value corresponding to the tag by multiplying the weight by a frequency value of the tag.

The determining of the recommendation product may include: equalizing the number of at least one review content of one review information among the distribution of the number of contents of each of the review information about the plurality of products, and each of the at least one review content of one review information. The method may further include calculating an RFS value of one product using the created time point and the at least one rating information.

Computing an evaluation value using the tag value of the tag corresponding to the searched product RFS value and the keyword of the searched product, and to determine the ranking of the plurality of products according to the evaluation value, and to recommend products corresponding to the higher ranking It may further comprise a step.

,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 빈도,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 개수,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 모든 단어들의 개수일 수 있다.Extracting, as a tag, a word having a frequency greater than or equal to a threshold value from at least one review content of one review information, calculates a frequency of each of the words in at least one review content of one review information using the following equation: Including the steps of:

,

Is the frequency of a particular word in at least one review content of one review information,

The number of specific words in at least one review content of one review information,

May be the number of all words in at least one review content of one review information.

,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 빈도,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 특정 단어의 개수,

는 하나의 리뷰 정보의 적어도 하나의 리뷰 컨텐츠에서의 모든 단어들의 개수일 수 있다.Computing the weight for the tag according to the degree of use of the tag in the review information for a plurality of products, the frequency of each of the words in the at least one review content of one review information using the following equation Calculating a;

,

Is the frequency of a particular word in at least one review content of one review information,

The number of specific words in at least one review content of one review information,

May be the number of all words in at least one review content of one review information.

The effects of the product recommendation server and the product recommendation method according to the present disclosure will be described as follows.

According to at least one of the embodiments of the present disclosure, there is an advantage that can recommend a product that meets the needs of the user.

According to at least one of the embodiments of the present disclosure, there is an advantage that the sales amount of the product can be increased.

1 is a block diagram illustrating a product recommendation system according to an exemplary embodiment.
2 is a block diagram illustrating a product recommendation server according to an exemplary embodiment.
3 is a flowchart illustrating a product recommendation method according to an exemplary embodiment.
4 are graphs illustrating performance evaluation of a product recommendation server and a product recommendation method according to an exemplary embodiment.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification and claims, when a part is said to "include" a certain component, it means that it can further include other components, without excluding other components unless specifically stated otherwise.

Hereinafter, a product recommendation server and a product recommendation method according to an embodiment will be described in detail with reference to necessary drawings.

In the present disclosure, a case where a product is recommended by the product recommendation system will be described as an example. However, the present invention is not limited thereto, and the product recommendation server and the product recommendation method according to the embodiment may be applied to recommending intangible content and services other than products.

1 is a block diagram illustrating a product recommendation system according to an exemplary embodiment. As shown, the product recommendation system includes a server 100 and user terminals 200 that can communicate with each other by a communication network 300.

The server 100 may provide a recommendation service that determines a ranking of a product according to a keyword input from each of the user terminals 200 and provides information on the ranking to the user terminals 200.

The user terminals 200 may access the server 100 and receive a recommendation service for a product. The user terminal 200 may be applied to various communication terminals generally known, such as a smartphone, a PC, a laptop.

The user may be connected to the server 100 through the communication network 300 to the user terminal 200 owned by the user, where the communication network 300 is a wired communication network 300 and a wireless communication network 300 as a general Internet line ) May include all. Furthermore, when the product recommendation method according to an embodiment is provided in a limited range such as a specific company, the network may be a virtual private network or an intra network accessible only to a limited area or an authorized person.

In addition, the server 100 may provide a recommendation service to the user terminals 200 by interworking with a general search service server. For example, the user may access a search service server that he or she normally uses to input a keyword to recommend a product. You can request In this case, the search service server may request a product recommendation according to the keyword of the user from the server 100, and the server 100 may provide a product recommendation result for the user to the search service server.

Next, the product recommendation server 100 according to the embodiment will be described with reference to FIG. 2.

2 is a block diagram illustrating a product recommendation server 100 according to an exemplary embodiment.

As shown, the product recommendation server 100 may include a review database 110, a controller 120, a review analysis database 130, and a communication unit 140.

The review database 100 may store information about products. The information about the products may include review information of each product. Here, one piece of review information about a product may include at least one of contents and ratings input by buyers for the product.

The controller 120 may include a tag extractor 122, a review analyzer 124, and a ranking determiner 126.

The tag extractor 122 may read review information about products to be recommended from the review database 110.

The tag extractor 122 extracts tags from each review information read from the review database 110. For example, the tag extractor 122 may extract a word frequently appearing in contents of all review information of a product as a tag having explanatory power for the product.

Specifically, when the contents of all review information corresponding to a specific product are referred to as one document d , the word frequency value TF for a specific word t in the document d is determined by Equation 1 below. do.

This makes sense as the ratio obtained by dividing the sum of the frequency of the particular word (t) all the words (t ') extraction of the frequency, in the document (d) extracted from the document (d).

The tag extractor 122 extracts words based on common nouns and proper nouns from the review information, and tags words of a product ( TAGd ) whose upper cumulative sum of the word frequency value tf is greater than or equal to a threshold among the extracted words. It can be extracted as an element of.

In addition, the tag extractor 122 may calculate a weight for each tag TAGd . For example, the words that appear frequently in all products are not suitable to describe a particular product, so to exclude such words, weights for the extracted tags can be calculated.

The weight IDF may be calculated using Equation 2 below.

는 특정 단어( d )를 포함하는 문서의 개수이고,

는 전체 문서의 개수이다. here

Is the number of documents that contain a specific word ( d ),

Is the total number of documents.

The tag extractor 122 may calculate a tag value TagValue by multiplying the weight IDF by the word frequency value TF of each tag TAGd as shown in Equation 3 below.

Where w is an element of a tag ( TAGd ). TagValue is a value that depends on the combination of tag and product. That is, even if the same tag, the tag value (TagValue) may be different according to the product. The higher the TagValue, the better the tag describes the product.

Next, the review analyzing unit 124 may also read review information about products that are to be recommended from the review database 110. The review analyzer 124 may analyze the review information by dividing the review information into three indicators (RFS: Recency, Frequency, Score).

(1) Recency

The review analyzer 124 may assign higher weights as the review information is recently written. For example, the review analyzer 124 may assign a weighted recency (WR) to each review information as shown in Table 1 below.

(2) Frequency

The review analyzer 124 may give a higher weight as the number of reviews of the corresponding product increases in consideration of the distribution of the number of reviews of the entire product. For example, when the review analysis unit 124 sorts the number of reviews of each product in ascending order, the weight for each product as shown in Table 2 below, according to the ranking belonging to the number of reviews of the product WF: Weighted Frequency can be assigned.

For example, the review analyzing unit 124 has three reviews of product 1, four reviews of product 2, ten reviews of product 3, and eight reviews of product 4 In this case, the weight WF of the product 1 may be 1.0, the weight WF of the product 2 is 1.1, the weight WF of the product 4 is 1.2, and the weight WF of the product 3 is 1.3.

(3) Score

The review analyzer 124 may read the rating included in the review information.

Then, the review analyzer 124 may calculate an RFS value of one product having n review information as shown in Equation 4 below.

The RFS value has a unique value for each product, and a higher value may mean that the corresponding product is more popular.

The review analysis database 130 may store tag information and RFS values corresponding to each product. Here, the tag information includes at least one tag corresponding to one product and a tag value of the tag.

Next, the ranking determination unit 126 may search for a product including a keyword input from the review analysis database 130 as a tag, and determine the ranking of the searched products. The ranking determination unit 126 may output information about products within a predetermined rank (for example, within 5th rank) to the communication unit 140 according to the ranking of each evaluation value of the products.

The ranking determination unit 126 may calculate an evaluation value TRFS for one product by using Equation 5 below.

Here, TagValue is a tag value calculated for the tag of one product including the input keyword as a tag, and RFS is the RFS value of the product.

In addition, when a plurality of keywords is input, the ranking determination unit 126 may calculate a total evaluation value (Total TRFS) for reflecting all the plurality of input keywords using Equation 6 below.

Where keyword is the set of words entered by the user.

The ranking determination unit 126 may output information about products having a threshold rank or higher to the communication unit 140 according to the ranking according to the total evaluation value Total TRFS.

For example, when the product recommendation server 100 receives a plurality of keywords, the product recommendation server 100 may calculate the total evaluation value and output information on the top 5 products having the largest value to the communication unit 140.

The communication unit 140 enables communication between the product recommendation server 100 and the communication network 300, between the product recommendation server 100 and another search server, or between the product recommendation server 100 and the user terminal 200. It may include one or more modules.

Next, with reference to FIG. 3, the product recommendation method is demonstrated.

3 is a flowchart illustrating a product recommendation method according to an exemplary embodiment.

First, the tag extractor 122 extracts tags from each review information read from the review database 110 (S100). In addition, the tag extractor 122 may calculate a weight for each tag. The tag extractor 122 may calculate a tag value TagValue by multiplying the weight IDF by the word frequency value TF of each tag TAGd using Equation 3 above.

Next, the review analyzing unit 124 divides the review information into three indicators (RFS) and analyzes (S110). The review analyzer 124 may calculate RFS values for each product by using Equation 4 above.

Then, the review analysis database 130 may store the tag information extracted by the tag extractor 122 and the RFS value calculated by the review analyzer 124 in correspondence with each product.

Next, the ranking determination unit 126 searches for a product including a keyword input from the review analysis database 130 as a tag, and determines the ranking of the searched products (S140).

The ranking determination unit 126 provides information about products having a threshold rank or higher as a recommendation product list according to the ranking of each evaluation value of the products (S150).

According to the above-described embodiment, the product recommendation server 100 and the product recommendation method extract a tag of the product, calculate a tag value and an RFS value, and provide information on a product that matches the input keyword. The advantage is that you can recommend products that meet your needs.

The performance of the product recommendation server 100 and the product recommendation method may be evaluated by Equation 7 below.

Here, G [i] may be a tag value (TagValue) of the i- th product.

Equation (7) refers to the DCG (Discounted Cumulative Gain) scale, and the value becomes smaller by log _b i as the lower rank increases, so that the DCG value increases as the highly related products appear in the higher rank. That is, the higher the i- th product is related to the search result, the better the higher ranking means.

In addition, the performance of the product recommendation server 100 and the product recommendation method can be evaluated by variance of the TRFS value. The smaller the TRFS variance value, the more secure the ranking by the particular tag. When multiple keywords are input from the user, the TRFS variance must be small to reflect the effects of the keywords evenly.

In order to evaluate the performance, the DCG value and the TRFS variance value when the TRFS value is calculated by other equations (Equations 8 to 10 below) in addition to the equation 5 are evaluated.

18,525 reviews of 658 books on management / economics were used for performance evaluation. 6,028 tags were extracted from 658 books, 200 of which were sampled for the experiment. For each tag, the top five books with TRFS values by functions according to Equations 5 and 8 to 10 are configured as recommended items. The average DCG of the recommendation result obtained by applying 200 tags to four functions is as shown in FIG. 4 (a), and the average of the TRFS variance is as shown in FIG. 4 (b).

4 are graphs illustrating performance evaluation of a product recommendation server and a product recommendation method according to an exemplary embodiment.

As shown in (a) of FIG. 4, it can be seen that the DCG of the function according to Equation 5 is 10.845, which is the largest. Also, as shown in FIG. 4B, the TRFS variance of the function according to Equation 5 is the smallest as 46.84. Therefore, when using the function according to equation (5), it can be seen that the performance of the product recommendation method is better than other functions.

Embodiments of the present invention are not implemented only through the above-described apparatus and / or method, and in order to realize a function corresponding to the configuration of the embodiments of the present invention, a program recorded on a recording medium or a recording medium on which the program is recorded is provided. The implementation may be easily implemented by those skilled in the art to which the present invention pertains from the description of the above-described embodiments.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (14)

Communication,
A review database that stores review information for multiple products, and
A controller electrically connected to the communication unit and the review database
Including
One review information corresponding to one product includes at least one review content and at least one rating information,
The controller is a product including a keyword received through the communication unit of the plurality of products as a tag, according to the evaluation value calculated using the latestness, frequency, and rating of the review information corresponding to each of the plurality of products When listing the plurality of products in ranking order, recommend products within a critical ranking,
The controller extracts, as a tag, a word having a frequency equal to or greater than a threshold value from at least one review content of the one review information, and weights the tag according to the degree of use of the tag in review information about the plurality of products. Calculating a tag value corresponding to the tag by multiplying the weight value by a frequency value of the tag,
The controller may be configured to determine the equality of the number of the at least one review content of the one review information and the at least one review content of the one review information among the distribution of the number of contents of each of the review information about the plurality of products. Calculate an RFS value of the one item using a time point and the at least one rating information,
The controller searches for a product having a tag corresponding to the keyword among the plurality of products, calculates an evaluation value by using an RFS value of the searched product and a tag value of a tag corresponding to the keyword of the searched product. Determine a ranking of the plurality of products according to the evaluation value, recommend products corresponding to a higher ranking,
The controller calculates an RFS value of the one product by using the following equation,

,
The WF is a weight determined according to the equality of the number of at least one review content of the one review information among the distribution of the number of contents of each of the review information for the plurality of products, and the WR is at least one of the one review information. A weight determined according to a time point at which one piece of review content is written, and S is a rating corresponding to each of at least one piece of review information of the one piece of review information,
The control unit calculates the evaluation value by multiplying a tag value of a tag corresponding to the keyword of the searched product by a square root of an RFS value of the searched product.
Product recommendation server. delete delete delete The method of claim 1,
The controller calculates a frequency of each of words in at least one review content of the one review information using the following equation,

,

Is the frequency of a particular word in at least one review content of said one review information,

Is the number of the specific words in at least one review content of the one review information,

Is the number of all words in at least one review content of the one review information,
Product recommendation server. The method of claim 5,
The controller calculates a weight for the tag using the following equation,

,

Is the weight for the tag,

Is the total number of review information for the plurality of products,

Is the number of review information including the specific word among review information about the plurality of products,
Product recommendation server. delete delete Reading review information for a plurality of products from a review database, wherein one review information corresponding to one product includes at least one review content and at least one rating information;
Receiving a keyword through a communication unit,
Search for a product including the keyword as a tag among the plurality of products, and the searched products in order of ranking according to an evaluation value calculated using the latestness, frequency, and rating of review information corresponding to each of the searched products. Determining products that are within the threshold ranking as recommended products when listing them
Including,
Determining as the recommended product,
Extracting, as a tag, a word having a frequency greater than or equal to a threshold value from at least one review content of the one review information,
Calculating a weight for the tag according to the degree to which the tag is used in review information for the plurality of products, and
Calculating a tag value corresponding to the tag by multiplying the weight value by the weight value of the tag,
Determining as the recommended product,
An equality of the number of at least one review content of the one review information among distributions of the number of contents of each of the review information for the plurality of products, a time point at which each of the at least one review content of the one review information is created, and Calculating an RFS value of the one product using the at least one rating information, and
Search for a product having a tag corresponding to the keyword among the plurality of products, calculate an evaluation value using an RFS value of the searched product and a tag value of a tag corresponding to the keyword of the searched product, and evaluate the evaluation Determining a ranking of the plurality of products according to a value, and recommending products corresponding to a higher ranking;
The RFS value of the one product is calculated using the following equation,

,
The WF is a weight determined according to the equality of the number of at least one review content of the one review information among the distribution of the number of contents of each of the review information for the plurality of products, and the WR is at least one of the one review information. A weight determined according to a time point at which one piece of review content is written, and S is a rating corresponding to each of at least one piece of review information of the one piece of review information,
The calculating of the evaluation value includes calculating the evaluation value by multiplying a tag value of a tag corresponding to the keyword of the searched product by a square root of an RFS value of the searched product.
How to recommend a product. delete delete delete The method of claim 9,
Extracting, as a tag, a word having a frequency greater than or equal to a threshold value from at least one review content of the one review information,
Calculating the frequency of each of the words in the at least one review content of the one review information using the equation

,

Is the frequency of a particular word in at least one review content of said one review information,

Is the number of the specific words in at least one review content of the one review information,

Is the number of all words in at least one review content of the one review information,
How to recommend a product. The method of claim 9,
Computing the weight for the tag according to the degree of use of the tag in the review information for the plurality of products,
Calculating the frequency of each of the words in the at least one review content of the one review information using the equation

,

Is the frequency of a particular word in at least one review content of said one review information,

Is the number of the specific words in at least one review content of the one review information,

Is the number of all words in at least one review content of the one review information,
How to recommend a product.

Images