KR102174206B1 - Method and apparatus for online product recommendation considering reliability of product - Google Patents

Abstract

Disclosed is a method and apparatus for recommending online products in consideration of product reliability. An online product recommendation method according to an embodiment includes the steps of determining a product preference of the user for the first products based on the user's online behavior with respect to the first products, and determining another product preference for the second products. And determining product reliability for the second products based on users' online behavior, and providing a product recommendation list to the user based on the product preference and the product reliability.

Description

Online product recommendation method and device considering product reliability {METHOD AND APPARATUS FOR ONLINE PRODUCT RECOMMENDATION CONSIDERING RELIABILITY OF PRODUCT}

The following embodiments relate to a method and apparatus for recommending an online product in consideration of the reliability of the product.

Consumers can easily purchase products of various items due to the recent activation of online shopping malls. Sellers can also easily promote their products online, allowing them to realize profits in various forms.

Consumers check product information in an online shopping mall or use product purchase evaluation information of other consumers to purchase a desired product. Consumers can make various choices for product purchase through a wide range of information, but on the contrary, consumers have to judge whether or not the vast amount of information is reliable information, so there are many cases where consumers do not have certainty about their product selection. .

For example, products with a lot of product information but few purchase reviews often do not purchase products because consumers think that there are few other consumers who have actually used the product, or hesitate to purchase in product selection.

On the contrary, among the products that have many reviews of purchases of specific products in shopping malls and received good reviews, there are cases where they contain malicious information that makes consumers believe that the product is a good product through false product purchase reviews.

In order to provide meaningful and reliable information to consumers in online shopping malls, a more objective and evaluable form of product reliability evaluation method is needed.

Embodiments may provide a technology for recommending a product having an objective product reliability to a user by reflecting the user's product preference and product reliability derived from other users in the product recommendation step.

An online product recommendation method according to an embodiment includes the steps of determining a product preference of the user for the first products based on the user's online behavior with respect to the first products, and determining another product preference for the second products. And determining product reliability for the second products based on users' online behavior, and providing a product recommendation list to the user based on the product preference and the product reliability.

The online behavior of the user includes the user's product evaluation score, the number of product clicks, whether or not to steam the product for future purchase of the first products, and the time spent on the product page, and the online behavior of the other users , The number of product mentions by the other users, product evaluation scores, product clicks, product wishes for future purchase of the second products, time spent on product page, and number of times stayed on product page.

The determining of the product preference includes extracting an action value for each online action performed on each of the first products, and calculating an action weight for each online action based on the action value. And calculating a current product preference of the user for the first products based on the behavior value and the behavior weight, and the current product preference and the user's previous product for the first products It may include calculating the product preference based on the preference.

The calculating of the action weight includes: normalizing the action value, calculating an execution rate of each online action using the normalized action value, and the performance rate and the second evaluation by the user. 1 It may include calculating the action weight based on the number of products.

The calculating of the product preference may include calculating a time weight for the first products based on the current product preference and the previous product preference, and reflecting the time weight to the current product preference to obtain the product preference. It may include the step of calculating.

The determining of the product reliability may include the other users for the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of stays on the product page. It may include calculating the degree of interest of the people, and calculating the product reliability based on the number of mentions of the product and the degree of interest.

The calculating of the action weight includes: normalizing the action value, calculating an execution rate of each online action using the normalized action value, and the performance rate and the second evaluation by the user. 1 It may include calculating the action weight based on the number of products.

The calculating of the product preference may include calculating a time weight for the first products based on the current product preference and the previous product preference, and reflecting the time weight to the current product preference to obtain the product preference. It may include the step of calculating.

The determining of the product reliability may include the other users for the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of stays on the product page. It may include calculating the degree of interest of the people, and calculating the product reliability based on the number of mentions of the product and the degree of interest.

Providing the product recommendation list may include comparing product preferences of the other users for the first products and product preferences of the user for the first products, and comparing the product preferences of the user among the other users. Extracting users with similar product preferences; predicting the user's product preference for the third products based on the extracted user product preferences for third products; and the third product And generating a product list by sorting them according to predicted product preferences, and providing the product recommendation list based on the product list and the product reliability.

The extracting may include selecting a reference product from among the first products to compare product preferences with other users, extracting a first product preference of the user with respect to the reference product, and the reference Extracting a second product preference of the other users for a product, comparing the first product preference with the second product preference, and similar to the first product preference among the other users based on the comparison result. It may include extracting users as users with product preferences similar to the product preferences of the user.

Providing the product recommendation list may include comparing product preferences of the other users for the first products and product preferences of the user for the first products, and comparing the product preferences of the user among the other users. Extracting users with similar product preferences; predicting the user's product preference for the third products based on the extracted user product preferences for third products; and the third product And generating a product list by sorting them according to predicted product preferences, and providing the product recommendation list based on the product list and the product reliability.

The extracting may include selecting a reference product from among the first products to compare product preferences with other users, extracting a first product preference of the user with respect to the reference product, and the reference Extracting a second product preference of the other users for a product, comparing the first product preference with the second product preference, and similar to the first product preference among the other users based on the comparison result. It may include extracting users as users with product preferences similar to the product preferences of the user.

The predicting of the product preference may include extracting the user's product preference for the product selected by the user's action on the product from among the first products, and the user's product preference for the selected product. Calculating a similarity between the user and the extracted users based on product preferences of the extracted users with respect to the selected product, and the user's product with respect to the third products based on the similarity It may include the step of predicting the preference.

The providing of the product recommendation list includes calculating a ranking value indicating a recommendation ranking for products in the product list based on the predicted product preference and the product reliability, and according to the ranking value. It may include the step of providing the product recommendation list.

An online product recommendation apparatus according to an embodiment includes a preference determiner for determining a product preference of the user for the first products based on the user's online behavior with respect to the first products, and And a reliability determiner for determining product reliability for the second products based on the online behavior of other users, and a recommender for providing a product recommendation list to the user based on the product preference and the product reliability.

The online behavior of the user includes the user's product evaluation score, the number of product clicks, whether or not to steam the product for future purchase of the first products, and the time spent on the product page, and the online behavior of the other users , The number of product mentions by the other users, product evaluation scores, product clicks, product wishes for future purchase of the second products, time spent on product page, and number of times stayed on product page.

The preference determiner extracts an action value for each online action performed on each of the first products, calculates an action weight for each online action based on the action value, and calculates the action value and The user's current product preference for the first products may be calculated based on the action weight, and the product preference may be calculated based on the current product preference and the user's previous product preference for the first products. have.

The preference determiner normalizes the behavior value, calculates the execution ratio of each online behavior using the normalized behavior value, and based on the execution ratio and the number of the first products evaluated by the user, the Behavior weights can be calculated.

The preference determiner may calculate the product preference by calculating a time weight for the first products based on the current product preference and the previous product preference, and reflecting the time weight to the current product preference.

The reliability determiner calculates the degree of interest of the other users with respect to the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of stays on the product page And, the product reliability may be calculated based on the number of times the product is mentioned and the degree of interest.

The preference determiner normalizes the behavior value, calculates the execution ratio of each online behavior using the normalized behavior value, and based on the execution ratio and the number of the first products evaluated by the user, the Behavior weights can be calculated.

The preference determiner may calculate the product preference by calculating a time weight for the first products based on the current product preference and the previous product preference, and reflecting the time weight to the current product preference.

The reliability determiner calculates the degree of interest of the other users with respect to the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of stays on the product page And, the product reliability may be calculated based on the number of times the product is mentioned and the degree of interest.

The recommender compares the product preferences of the other users for the first products and the product preferences of the user for the first products, and among the other users, a user having a product preference similar to the product preference of the user Are extracted, the user's product preference for the third products is predicted based on the extracted product preferences of the users for the third products, and the third products are sorted according to the predicted product preference A list may be generated, and the product recommendation list may be provided based on the product list and the product reliability.

The recommender selects a reference product from among the first products to compare product preferences with the other users, extracts a first product preference of the user with respect to the reference product, and the other user with respect to the reference product Extracts the second product preference of the user, compares the first product preference with the second product preference, and determines users similar to the first product preference among the other users based on the comparison result. Users with similar product preferences can be extracted.

The recommender extracts the user's product preference for the product selected by the user's action on the product from among the first products, and the user's product preference for the selected product and the extraction for the selected product A similarity of the product propensity of the user and the extracted users may be calculated based on the product preferences of the users, and the product preference of the user with respect to the third products may be predicted based on the similarity.

The recommender may calculate a ranking value indicating a recommendation ranking for products in the product list based on the predicted product preference and the product reliability, and provide the product recommendation list according to the ranking value.

1 is a diagram illustrating an example of a product recommendation method for promoting product sales in an existing online shopping mall.
2 is a block diagram schematically illustrating an online product recommendation apparatus according to an exemplary embodiment.
3 is a flowchart schematically showing the overall operation of the online product recommendation apparatus shown in FIG. 2.
FIG. 4 is a detailed flowchart illustrating a process of determining a product preference determined by the preference determiner illustrated in FIG. 3.
FIG. 5 is a detailed flowchart illustrating a process of determining product reliability determined by the reliability determiner shown in FIG. 3.
6 is a flowchart illustrating a process of providing a product recommendation list provided by the recommender shown in FIG. 3 in detail.
FIG. 7 is a diagram illustrating an example in which a preference determiner extracts a user action shown in FIG. 4 as an action value.
FIG. 8 is a diagram illustrating an example in which the preference determiner normalizes the behavior value shown in FIG. 7.
9 is a diagram illustrating an example in which the preference determiner calculates the normalized action value shown in FIG. 8 as an execution ratio.
FIG. 10 is a diagram illustrating an example in which the preference determiner calculates an action weight based on the performance ratio shown in FIG. 9.
11 is a diagram illustrating an example in which a preference determiner calculates a current product preference based on the action weight shown in FIG. 10.
12 is a more detailed flowchart illustrating a process of determining the reliability of the product shown in FIG. 5 by the reliability determiner.
FIG. 13 is a flowchart sequentially listing all steps in which the online product recommendation device provides the product recommendation list shown in FIG. 3.
14 is a diagram illustrating an example in which a recommender calculates a predicted product preference for calculating a ranking value illustrated in FIG. 6.

Hereinafter, exemplary embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the rights of the patent application is not limited or limited by these embodiments. It should be understood that all changes, equivalents, or substitutes to the embodiments are included in the scope of the rights.

The terms used in the examples are used for illustrative purposes only and should not be interpreted as limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present specification, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the presence or addition of elements or numbers, steps, actions, components, parts, or combinations thereof, does not preclude in advance.

Terms such as first or second may be used to describe various components, but the components should not be limited by terms. The terms are only for the purpose of distinguishing one component from other components, for example, without departing from the scope of rights according to the concept of the embodiment, the first component may be named as the second component, and similarly The second component may also be referred to as a first component.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms as defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this application. Does not.

In addition, in the description with reference to the accompanying drawings, the same reference numerals are assigned to the same components regardless of the reference numerals, and redundant descriptions thereof will be omitted. In describing the embodiments, when it is determined that a detailed description of related known technologies may unnecessarily obscure the subject matter of the embodiments, the detailed description thereof will be omitted.

1 is a diagram illustrating an example of a product recommendation method for promoting product sales in an existing online shopping mall.

Referring to FIG. 1, the online shopping mall may provide a personalized product recommendation service through analysis of consumer purchase history, comments, and/or product distribution information to promote product advertisement and product sales.

For example, in the case of A online shopping mall, consumer activity information and use during the shopping process, such as classification of products viewed by consumers, number of product information inquiry, product information sharing, number of question registrations, final payment price, and actual purchase. By analyzing patterns in real time, you can recommend related products to consumers. For another example, in the case of the B online shopping mall, a consumer's consumption pattern may be analyzed through the consumer's product purchase history, and related products may be recommended when a consumer purchases a product. In addition, in the online shopping mall B, a'book match' service that compares similarities between products and reflects the similarities in product recommendations may be provided.

The pattern-based product recommendation method is a personalized product recommendation method, and as shown in FIG. 1, by analyzing a purchase behavior pattern of consumers, an item purchased by another consumer having a similar purchase pattern may be recommended to the corresponding consumer. When a product is recommended to a consumer through a pattern-based product recommendation method, since the product is recommended only with the consumer's purchasing behavior pattern, there may be cases in which the consumer is recommended a product that he or she does not intend to purchase.

The propensity-based recommendation method analyzes the product purchased by the consumer and the consumer's profile and extracts the consumer's product propensity to recommend a product to the consumer based on the consumer's product propensity. Since the propensity-based recommendation method recommends products based on the consumer's product propensity, products that consumers would like are usually recommended, but products that are not evaluated by consumers are very unlikely to be created as recommended products, so consumers are fixed products. Only the list can be recommended.

When comparing the case of sending direct mail (DM) for all product information discounted to consumers by the online shopping mall and the case of sending personalized product information DM through analysis of the consumer's pattern and/or disposition, the consumer's pattern And/or when personalized product information is provided through propensity analysis, the purchase conversion rate of consumers may increase by 10-20%. As can be seen from these examples, a personalized recommendation function may be required to improve the purchase rate in an online shopping mall.

In order to secure competitiveness and increase sales of online shopping malls, it may be important to discover potential customers who do not have an active purchase history. In addition, companies can reinforce their corporate competitiveness by implementing differentiated one-on-one marketing for each consumer and customer relationship management strategies that understand consumers from the perspective of consumers and strengthen relationships with consumers.

The personalized product recommendation system may help consumers make decisions or improve consumers' product purchasing rates by recommending information and products suitable for each consumer's individual interest. That is, the recommendation system automatically recommends and provides information and items tailored to each individual's taste, thereby reducing time and cost on the consumer side and increasing profits on the corporate side.

The personalized product recommendation method is divided into content-based recommendation and collaborative filtering.

Content-based recommendation is a method of finding content with the highest similarity to customer information by normalizing product attributes to arbitrary values or characters, and recommendation can be simple and fast. However, since content-based recommendation is a simple form of recommending using content and personal information, only consistent recommendations can be made without considering personal preferences.

Collaborative filtering-based recommendation may be a method of recommending items that new users think will be of interest based on consumer preferences. Collaborative filtering is the most widely used information filtering technique to build a recommendation system, and it can be used most academically and commercially.

In addition, the item-based recommendation method may be a technique of recommending items that are similar to items purchased by similar consumers through calculation of similarity between items. However, in item-based recommendation, a fixed recommendation list can be created because the similarity between items calculated once is less likely to change.

In order to solve the limitation of the above-described product recommendation method, a user-based recommendation method for generating various recommendation lists by comparing similarities between users who have lowered the rating may be studied. However, the user-based recommendation method may have a cold start problem that cannot be recommended when the rating history or number of items is insufficient.

The hybrid recommendation method is a recommendation method that combines item-based similarity and user-based similarity in order to solve the phenomenon of item shortage such as a cold start problem. Although the hybrid recommendation method partially solves the limitations such as the cold start problem, the computational complexity may be relatively high.

In addition, it may not be possible to confirm whether the recommended product is actually a reliable product using the above-described recommendation method. If a specific group indiscriminately gives a high product review score, the probability of recommending the product may increase.

In order to solve this limitation, the embodiments reflect the product reliability that can know whether the recommended product is a reliable product when the product is selected as a candidate set of the product recommendation list in the product recommendation step, and recommend products with low product reliability. It can provide technology that can be removed from.

Hereinafter, embodiments will be described with reference to FIGS. 2 to 14.

FIG. 2 is a block diagram schematically illustrating an apparatus for recommending an online product according to an exemplary embodiment, and FIG. 3 is a flowchart schematically illustrating an overall operation of the apparatus for recommending an online product shown in FIG. 2.

Referring to FIG. 2, the online product recommendation device 10 includes a preference determiner 100, a reliability determiner 200, and a recommender 300.

The online product recommendation device 10 may provide a personalized product recommendation list reflecting product reliability to the user. The user may be a consumer using an online shopping mall or the like.

The online product recommendation device 10 may analyze a user's consumption pattern to determine a user's disposition and perform a recommendation based on this. Product recommendation performed by the online product recommendation device 10 may utilize a user-based collaborative filtering technique. When the online product recommendation device 10 provides a product recommendation list, a user may recommend a product that has not been previously purchased.

The preference determiner 100 may determine a user's product preference for the first products based on the user's online behavior with respect to the first products.

The first products may be products that are subject to an action performed by a user online. The user's online behavior may include the user's product evaluation score, the number of product clicks, whether or not to steam a product for future purchase of the first products, and a time spent on a product page.

The preference determiner 100 may transmit the product preference to the recommender 300.

The reliability determiner 200 may determine the reliability of the second products based on the online behavior of other users with respect to the second products.

The second products may be products that are subject to an action performed by other users online. The second products may include first products. The online behavior of other users includes the number of products mentioned by other users, the product evaluation score, which is the popularity of the product in the online shopping mall, the number of product clicks, the number of wishes for future purchases of the second products, the time spent on the product page, and the product. This may include the number of times you stayed on the page.

The reliability determiner 200 may transmit the product reliability to the recommender 300.

The recommender 300 may provide a product recommendation list to a user based on product preference and product reliability. The products in the product recommendation list may not include the first products.

3, in the online product recommendation device 10, the preference determiner 100 determines a product preference (310), the reliability determiner 200 determines the product reliability (320), and the recommender 300 A product recommendation list may be provided to a user by providing 330 a product recommendation list based on product preference and product reliability.

FIG. 4 is a detailed flowchart illustrating a process of determining a product preference determined by the preference determiner illustrated in FIG. 3.

4, the preference determiner 100 may determine an action weight and a time weight based on the user action 311 (312, 313), and calculate a product preference based on the action weight and/or time weight. There is (314).

The preference determiner 100 may extract an action value for each online user's action performed on each of the first products. An example in which the preference determiner 100 extracts an action value will be described in detail with reference to FIG. 7.

The preference determiner 100 may calculate an action weight for each online action based on the action value. More specifically, the preference determiner 100 may normalize the behavior value. The preference determiner 100 may calculate the performance ratio of each online action using the normalized action value. The preference determiner 100 may calculate an action weight based on the performance ratio and the number of first products evaluated by the user. An example in which the preference determiner 100 calculates an action weight based on an action value will be described in detail with reference to FIGS. 8 to 10.

The preference determiner 100 may calculate a user's current product preference for the first products based on the action value and the action weight. More specifically, the preference determiner 100 may calculate a product preference based on a current product preference and a user's previous product preference for the first products. The preference determiner 100 may calculate time weights for the first products based on the current product preference and the previous product preference (313). The preference determiner 100 may calculate the product preference by reflecting the time weight on the current product preference (314). An example in which the preference determiner 100 calculates a product preference based on an action weight will be described in detail with reference to FIG. 11.

FIG. 5 is a detailed flowchart illustrating a process of determining product reliability determined by the reliability determiner shown in FIG. 3.

5, the reliability determiner 200 can derive the popularity of the service and the amount of SNS mention of the product based on the actions 321 of other users (322, 323), and the popularity of the service and the product Product reliability may be determined based on the amount of SNS mention (324).

The reliability determiner 200 may calculate the degree of interest of other users with respect to the second products based on the product evaluation scores of other users, the number of wishes, the time spent on the product page, and the number of times they stayed on the product page. The level of interest may be the level of popularity in the service and/or the amount of SNS mention of a product. The reliability determiner 200 may calculate product reliability based on the number of product mentions and interest. An example in which the reliability determiner 200 determines product reliability will be described in detail in FIG. 12.

6 is a flowchart illustrating a process of providing a product recommendation list provided by the recommender shown in FIG. 3 in detail.

6, the recommender 300 extracts a similar user set based on the product preference transmitted from the preference determiner 100 and the product reliability transmitted from the reliability determiner 200 (331), and Predict product preference for third products based on product preference (332), generate a product list according to the predicted product preference (333), and provide a product recommendation list based on the product list and product reliability. Yes (334).

The recommender 300 compares product preferences for first products of other users and product preferences for first products of the user, and extracts users with product preferences similar to the user's product preferences among other users. have. The recommender 300 may select a reference product from among the first products in order to compare product preferences with other users.

The recommender 300 may extract the user's first product preference with respect to the reference product. The recommender 300 may extract second product preferences of other users with respect to the reference product.

The recommender 300 may compare the first product preference and the second product preference. The recommender 300 may extract users similar to the first product preference among other users as users with product preference similar to the user's product preference, based on the comparison result of the first product preference and the second product preference. have.

The recommender 300 may predict the user's product preference for the third products based on the extracted user's product preferences for the third products. The third products may be products that do not include the first products among the second products.

The recommender 300 may extract a user's product preference for a selected product based on the user's action on the product from among the first products. The recommender 300 may calculate the similarity between the user and the extracted users based on the user's product preference for the selected product and the extracted users' product preference for the selected product. The recommender 300 may predict a user's product preference for third products based on the similarity.

The recommender 300 may generate a product list by sorting the third products according to the predicted product preference.

The recommender 300 may provide a product recommendation list based on a product list and product reliability. More specifically, the recommender 300 may calculate a ranking value indicating a recommendation ranking for products in a product list based on the predicted product preference and product reliability. The recommender 300 may provide a product recommendation list according to the ranking value. An example in which the recommender 300 provides a product recommendation list will be described in detail with reference to FIGS. 13 to 14.

FIG. 7 is a diagram illustrating an example in which the preference determiner extracts a user action shown in FIG. 4 as an action value, and FIG. 8 is a diagram illustrating an example in which the preference determiner normalizes the action value shown in FIG. 7.

Referring to FIG. 7, when a user performs an action on a product, the preference determiner 100 may extract a value for each action as shown in FIG. 7. USER_ID, ITEM_ID, RATING, CLICK_NUM, IS_LIKE, and STATY_TIME are the user identification number, product identification number, evaluation score the user gave on the product, the number of clicks the user clicked on the product, whether the user wished to make a wish on the product, It can indicate the time spent on the product page after clicking a product.

Referring to FIG. 8, when the preference determiner 100 converts a user's action into a value, each action may have a different value range and even a different type of value may exist. For example, the rating is expressed between 0 and 5 points, the minimum value for the number of clicks is 0, and the maximum value may be different for each user. Because steaming has a Boolean value for whether steaming is done or not, the range of values may not exist like other actions. Therefore, in order for the preference determiner 100 to calculate the weight of an action, a normalization process may be required to convert the action value into a range of the same value.

Normalization of the behavior value is calculated through Equation 1.

에서 u는 사용자의 식별 번호, i는 상품 식별 번호, b는 사용자가 수행한 행위의 형태일 수 있다. 사용자(u)가 상품(i)에 대한 행위(b)에 대한 행위 값을 정규화 함을 의미할 수 있다.

와

는 사용자(u)가 수행한 행위(b)에 대한 최대값과 최소값을 나타낼 수 있다.Normalized value

In, u may be a user identification number, i may be a product identification number, and b may be a form of an action performed by the user. It may mean that the user u normalizes the action value for the action b for the product i.

Wow

May represent the maximum and minimum values for the action (b) performed by the user (u).

For example, if there is a user who gives only 3 to 4 points in the evaluation score, the maximum and minimum values of each user's actions are used because the 4 points should be considered the most satisfactory value from the user's point of view.

Since the range of the action value is different for each user, the action value can be normalized to 0 to 1. When normalization is performed, all behavior values can be expressed as [0, 1].

는 사용자의 행위를 정규화하기 위한 실제 행위 값을 나타낸다. 즉,

는 사용자(u)가 상품(i)에 대해 수행한 행위(b)에 대한 실제 값일 수 있다. 평가 점수를 4점을 줬다면 4점이 그대로

에 대입될 수 있다.

Represents the actual behavior value to normalize the user's behavior. In other words,

May be an actual value for the action (b) performed by the user (u) on the product (i). If you gave the evaluation score 4 points, it will be 4 points.

Can be substituted for

An example in which the preference determiner normalizes the behavior value is shown in FIG. 8. 8 is a value obtained by normalizing behavior values for the three products of FIG. 7.

_ID is used as a unique identification number of a user and/or product. N(r) is a value obtained by normalizing the evaluation score given by the user to the product, and it is converted to 0.8, which is the normalized value (4-0)/(5-0)=4/5 by giving 4 points in FIG. In the same way, the remaining values can also be expressed as [0,1].

N(c) and N(s) may be values obtained by normalizing the number of clicks for the product and the time spent after the click. Since the maximum values of N(c) and N(s) are different for each user, they can be normalized separately for each user.

9 is a diagram illustrating an example in which the preference determiner calculates the normalized action value shown in FIG. 8 as an execution ratio.

Referring to FIG. 9, after normalizing an action value, an execution rate for each action may be calculated in order to calculate which action is performed a lot. The performance ratio can be calculated by Equation 2.

에서, 분모에서 사용되는 n은 행위의 개수이다. 즉, 수학식 2는 어떠한 상품(i)에 대해서 수행한 행위(b)에 대한 정규화 값을 동일 상품(i)에 대한 모든 행위 값을 모두 더한 값으로 나누어서 해당 행위(b)가 얼마만큼의 가중치를 가지는지를 계산하는 식일 수 있다.Performance rate

In, n used in the denominator is the number of actions. That is, Equation 2 divides the normalized value for the action (b) performed on a certain product (i) by the sum of all the action values for the same product (i), and how much weight the action (b) is It can be an expression that calculates whether it has

When the normalized action value of FIG. 8 is calculated as the execution rate, the action rate value as shown in FIG. 9 can be obtained. In FIG. 8, _ID is a unique identification number of a user and/or product, P(r) is a product evaluation score, P(c) is the number of product clicks, and P(s) is the time spent on the product page after clicking the product. It can be a ratio value.

For example, the P(r) of product 1 is the N(r) normalized value of product 1, 0.8, divided by the sum of the normalized values of all actions (0.8+0.5+1.0)=2.3, and the result is 0.34. I can. For the remaining values, the ratio value for all actions can be calculated through Equation 2.

10 is a diagram showing an example in which the preference determiner calculates an action weight based on the performance ratio shown in FIG. 9, and FIG. 11 is a view showing an example in which the preference determiner calculates a current product preference based on the action weight shown in FIG. A diagram showing an example.

Referring to FIG. 10, the preference determiner 100 may calculate an action weight indicating which action is important for each user. The preference determiner 100 may calculate a preference value for each product as a single value through the action weight value.

The action weight can be calculated by Equation 3.

에서, k는 사용자(u)가 평가한 총 상품 개수일 수 있다. 하나의 행위(b)에 대한 비율 값을 모두 더하여 평균화 한 값이 행위(b)에 대한 가중치 값일 수 있다. 예를 들어, 도 1에서 총 상품이 3개이기 때문에 k는 3이 될 수 있다.Behavior weight

In, k may be the total number of products evaluated by the user u. A value obtained by adding all ratio values for one action (b) and averaged may be a weight value for the action (b). For example, since the total number of products in FIG. 1 is 3, k may be 3.

U is a unique identification number of the user, and W(r) may be a weight value for an evaluation score action. The preference determiner 100 may calculate the sum of the ratio values P(r) (0.34+0.65+0.15) and divide this value by 3, which is k, to calculate a weight value for the final rating action.

W(c) may be the number of product clicks, W(s) may be the time spent on the product page after clicking the product, and W(l) may be the weight of the product wishing. Since steaming is a boolean value, it is difficult to normalize the steaming value or calculate the ratio value of the steaming value, so the action weight can be calculated using the steaming rate value as it is.

Referring to FIG. 11, a current product preference value for each product may be calculated through the calculated action weight. The current product preference value for each product can be calculated by summing all values obtained by multiplying the actual value for each action and the action weight value obtained through the above-described calculation process. However, since the range of values for the number of clicks or product preference is large, many fluctuations may occur according to the corresponding value, so a log value can be taken to control the influence of the value.

An example of calculating product preference is shown in FIG. 11. In order for user 234 to obtain a preference value for product 108, first, the value (4, 10, 3600, 1) for the existing action (product evaluation score, product clicks, time spent on product page, wishing product) is obtained, and each action The current product preference by multiplying the star weight values (0.38, 0.37, 0.22, 0.33) by the existing value and sum ((4*0.38 + log(10)*0.37 +log(3600)*0.22 + 1*0.33)=5.37) Value can be calculated.

A time weight can be calculated to reflect the change in the user's preference for the product over time in the calculated current product preference.

If the product preference value of FIG. 11 is checked, it can be seen that different product preference values for the same item 108 are indicated. In fact, if you check the behavior of the product, you can see that the middle rating was initially assigned, but the higher rating was assigned later. This can be seen as giving a higher rating as the product feels better by the user over time.

Product preference considering the time weight can be calculated by Equation 4.

은 사용자(u)가 상품(i)에 대한 행위 가중치를 고려하여 계산한 선호도 값이며 t는 시간 값일 수 있다. t는 시간에 대한 값을 계산하기 위해서 사용될 수 있다. 가장 최근의 선호도 값을 계산하기 위해서 가장 오래된 선호도 값

부터 그 다음 시간의 선호도 값

을 고려하여

을 계산하고, 이를 최근 시간 t까지 재귀적으로 계산할 수 있다.

Is a preference value calculated by the user u in consideration of the action weight for the product i, and t may be a time value. t can be used to calculate the value over time. The oldest preference value to calculate the most recent preference value

From the next time preference value

In consideration of

Can be calculated and calculated recursively up to the latest time t.

The product preference calculation method considering the time weight may be a method of calculating a final preference value by multiplying a ratio value of a difference between the preference value of the current time t and the preference value of the previous time t-1.

For example, if the previous product preference value is 1.92 and the current product preference value is 3.00, the time weight of 3.00*(3.00/1.92)=4.68 may be converted into a product preference value in consideration. Otherwise, it can be converted to a product preference value of 1.92*(1.92/3.00)=1.22.

Since the fact that the user has multiple preferences for the same product implies that a lot of the user's preferences are reflected in the product, the preference value can be further amplified or reduced to make the user's preferences clear. This can be very helpful in selecting a comparison product to use in product list recommendations.

12 is a more detailed flowchart illustrating a process of determining the reliability of the product shown in FIG. 5 by the reliability determiner.

Referring to FIG. 12, products included in a product list may not reflect product reliability for the product itself. Product reliability is a value of whether the product is actually a reliable product, and may be affected by the user's product satisfaction. Products that are frequently mentioned in SNS (Social Networking Service) have a high probability that users actually prefer them, which means that they are reliable products. Product reliability can be calculated from the popularity of the service and the amount of mention on SNS.

The reliability determiner 200 may collect SNS data in order to measure the reliability of the product, determine whether the product is frequently mentioned, and reflect it in the reliability calculation. In addition, the reliability determiner 200 also receives a high interest from other users if the product has a lot of steaming in the online shopping mall service, the number of times other users click the product and stay on the product page, or if the product has a high average rating. It can be considered as a product that has reliable evidence, and can be reflected in the reliability score.

라고 지칭할 수 있다.

는 SNS의 언급량을 의미할 수 있다.

는 수학식 5로 계산할 수 있다.The reliability determiner 200 may collect SNS data. The reliability determiner 200 may extract only information related to product information from the collected SNS data. The reliability determiner 200 may calculate TF-IDF (Term Frequency-Inverse Document Frequency) for each product using only information related to the product. In the TF-IDF calculation performed by the reliability determiner 200, a word dictionary may be constructed based on a product name and only corresponding words may be calculated. Calculated TF-IDF

Can be referred to as.

May mean the amount of mention of SNS.

Can be calculated by Equation 5.

에서, i는 특정 상품을 나타내며 TF-IDF는 불리언 기반의 TF를 사용하여 단어 출현 빈도가 많더라도 1 값을 가짐으로써 높은 값이 나오지 않게 조절할 수 있다.

In, i denotes a specific product, and TF-IDF uses a Boolean-based TF and has a value of 1 even if the frequency of occurrence of a word is high, so that a high value can be adjusted so as not to appear.

값을 계산할 수 있다. 서비스내의 인기도

는 수학식 6을 통해 계산할 수 있다.In order to calculate the popularity in the service, the reliability determiner 200 takes the number of wishes for each product, the number of stays on the product page, and the average rating value of the product.

Value can be calculated. Popularity within the service

Can be calculated through Equation 6.

에서,

는 상품(i)에 대한 사용자들의 찜하기 총 횟수이며

는 상품(i)에 대한 클릭 후 일정시간 머무른 총 횟수일 수 있다.

는 상품(i)에 대한 평균 평가 점수일 수 있다.Popularity within the service

in,

Is the total number of users' wishes for product (i)

May be the total number of times of staying for a certain time after clicking on the product (i).

May be an average evaluation score for product (i).

The reliability determiner 200 may analyze articles written by users in SNS services such as blogs, cafes, intellectuals, Twitter, and Facebook in order to determine the amount of mention in SNS.

와 서비스내의 인기도를 나타내는

값을 합산하여 최종적인 상품에 대한 상품 신뢰도인

를 계산할 수 있다. 상품 신뢰도 점수가 높은 상품일수록 추천 상품 목록의 상위 결과에 포함될 확률이 높아질 수 있다. 상품 신뢰도

는 수학식 7로 계산할 수 있다.Reliability determiner 200 is the amount of SNS mention

And the popularity of the service

By summing the values,

Can be calculated. The higher the product reliability score, the higher the probability of being included in the top result of the recommended product list. Product reliability

Can be calculated by Equation 7.

는 수학식 7의 가중치

를 통해 서비스내의 인기도와 SNS 언급량의 기여도를 조절할 수 있다.Product reliability

Is the weight of Equation 7

Through this, you can adjust the popularity of the service and the contribution of the amount of mentions to SNS.

FIG. 13 is a flowchart sequentially listing all steps in which the online product recommendation device provides the product recommendation list shown in FIG. 3, and FIG. 14 is a recommender calculating a predicted product preference for calculating the ranking value shown in FIG. It is a diagram showing an example of doing.

Referring to FIG. 13, the online product recommendation apparatus 10 may generate a matrix using a product preference 1310 in consideration of a user action weight and a time weight (User-Item Matrix 1320).

The online product recommendation device 10 may extract users having a product preference similar to a user to be recommended through a cosine similarity equation using the generated matrix (1330, 1340).

In order to calculate the similarity between the user and other users, the online product recommendation device 10 is used to reflect the n products with the highest product preference of the user, m products with the lowest product preference, and the product tendency of the recent user. You can select o products that have recently been rated as standard products. This is because calculating the similarity between users and other users based on the preferences of all products can be costly and time consuming. For example, if n, m, and o are each 3, similarity between users can be calculated based on a minimum of 3 to a maximum of 9 reference products.

The online product recommendation device 10 may predict a user's product preference for third products based on the similar user set 1340 (1340).

The predicted product preference can be calculated through Equation 8.

는 사용자(u)의 상품(i)에 대한 예측 상품 선호도일 수 있다. 식에서

는 사용자(u)의 평균 선호도를 나타내며,

은 사용자(u)와 다른 사용자(u’)간의 유사도 값일 수 있다. 즉, 사용자의 평균 선호도 값에서 유사 사용자 집합의 평균 선호도를 뺀 후 사용자와의 유사도를 곱하고 모든 유사도 수치를 합하여 나눈 값이 상품의 선호도 예측 값일 수 있다. 예측된 선호도 값이 클수록 해당 사용자에게 추천될 확률이 높아질 수 있다.

May be the predicted product preference of the user u for the product i. In expression

Represents the average preference of the user (u),

May be a similarity value between the user u and another user u'. That is, a value obtained by subtracting the average preference of a set of similar users from the average preference value of the user, multiplying the similarity with the user, and summing all similarity values, may be a product preference prediction value. The higher the predicted preference value, the higher the probability of being recommended to a corresponding user.

Referring to FIG. 14, an example in which the online product recommendation apparatus 10 calculates a predicted product preference based on a similarity is illustrated. The user U1 may extract a similar user group to calculate a predicted product preference for product Item2. In the example of FIG. 14, it is assumed that the similar user group is a single user U2, in which case the similarity between U1 and U2 is 0.84.

In order to calculate the predicted score of Item2, the average preference value of U1 can be calculated (2.74) and the average preference value of U2 (2.6) can be calculated. The value obtained by multiplying U2's preference for Item2 (0.1) by subtracting the average preference of U2 (2.6) by the similarity value with U2 (0.84) can be the numerator ((0.84*(0.1-2.6)))=-2.1 .). The denominator is the sum of the similarity of a set of similar users. Since there is only one U2 similar user, a value of 0.84 can be used.

The preference value of the set of similar users is -2.1/0.84=-2.5, and by adding a value to the average preference of U1 (2.74), a predicted product preference value of 0.24 can be calculated for Item2. It can be seen that the similar user U2's calculated preference for Item2 is 0.1, a very low value, and the predicted value is also 0.24, which is similarly low.

The online product recommendation device 10 may recommend k products to users in order from a higher value by reflecting the product reliability value to the predicted product preference value.

The online product recommendation device 10 may calculate a ranking value, which is a ranking value according to a predicted product preference value among the product list. The ranking value can be calculated by Equation 9.

는 앞에서 계산한 사용자(u)가 아직 접해보지 않은 상품(i) 즉 제3 상품들에 대한 예측 상품 선호도 값이며,

는 상품(i)에 대한 신뢰도 값이다.

Is the predicted product preference value for the product (i) that the user (u) calculated previously has not yet encountered, that is, third products,

Is the reliability value for product (i).

의 조절을 통해 신뢰도 및 예측 선호도 점수의 반영 정도를 조절한다.

와

의 최적 값은 성능 평가를 통해 도출되는 값을 사용할 수 있다.That is, Equation 9 may be an equation for obtaining a predicted value for final ranking based on a sum of a product reliability value and a predicted rating. weight

The degree of reflection of reliability and predicted preference scores is adjusted through the adjustment of

Wow

The optimal value of can be a value derived through performance evaluation.

The online product recommendation device 10 may provide a product recommendation list to the user according to the ranking value.

The embodiments may provide an apparatus and method for recommending personalized products in consideration of product reliability in order to secure competitiveness of an online shopping mall and increase sales. The online product recommendation device 10 may reflect the user's disposition and the reliability of the items in the recommendation. The online product recommendation apparatus 10 may use the user preference to determine a user similar to a user of a recommendation target, and include products consumed by similar users but not yet consumed by the product recommendation target. The online product recommendation device 10 may analyze a user's consumption pattern for customized recommendation, extract a propensity, and predict a product that the user will consume in the future based on similarities between users. The online product recommendation device 10 may extract the user's propensity based on an action represented by a consumption pattern performed by the user in the service (a rating, click, favorite, review, etc.). In addition, when there are multiple reviews and actions for the same product, the online product recommendation device 10 may assign a weight to an item suitable for the latest tendency in consideration of the weight over time. The online product recommendation device 10 may collect SNS information on the reliability of the product to determine whether the product is actually reliable, evaluate the reliability of the product, and reflect the evaluated reliability in the recommendation. In the final product recommendation, the online product recommendation device 10 may provide a product recommendation list by increasing the ranking of the recommendation as a product with high reliability and fits the propensity in consideration of the reliability of the analyzed product.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. , Or may be permanently or temporarily embodyed in a transmitted signal wave. The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

As described above, although the embodiments have been described by the limited drawings, a person of ordinary skill in the art can apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments and claims and equivalents fall within the scope of the following claims.

Claims (20)

Determining a product preference of the user for the first products based on the user's online behavior with respect to the first products;
Determining product reliability for the second products based on online behavior of other users with respect to the second products; And
Providing a product recommendation list to the user based on the product preference and the product reliability; Including,
The user's online behavior,
Including a product evaluation score of the user, the number of product clicks, whether a product is steamed for future purchase of the first products, and a time spent on the product page,
The online behavior of the other users,
Including the number of product mentions by the other users, product evaluation scores, product clicks, product wishes for future purchases of the second products, time spent on product page, and number of times stayed on product page
How to recommend products online.
The method of claim 1,
The step of determining the product preference,
Extracting an action value for each online action performed on each of the first products;
Calculating an action weight for each online action based on the action value;
Calculating a current product preference of the user for the first products based on the action value and the action weight; And
Calculating the product preference based on the current product preference and the user's previous product preference for the first products
Online product recommendation method comprising a.
The method of claim 2,
The step of calculating the action weight,
Normalizing the behavior value;
Calculating an execution rate of each online action using the normalized action value; And
Calculating the action weight based on the performance ratio and the number of the first products evaluated by the user
Online product recommendation method comprising a.
The method of claim 2,
The step of calculating the product preference,
Calculating time weights for the first products based on the current product preference and the previous product preference; And
Calculating the product preference by reflecting the time weight on the current product preference
Online product recommendation method comprising a.
The method of claim 1,
The step of determining the product reliability,
Calculating a degree of interest of the other users with respect to the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of times they stayed on the product page; And
Calculating the product reliability based on the number of product mentions and the interest level
Online product recommendation method comprising a.
The method of claim 1,
Providing the product recommendation list,
Comparing the product preferences for the first products of the other users with the product preferences for the first products of the user, and extracting users with product preferences similar to the product preferences of the user from among the other users ;
Predicting the user's product preference for the third products based on the extracted user product preferences for the third products;
Generating a product list by sorting the third products according to predicted product preferences; And
Providing the product recommendation list based on the product list and the product reliability
Online product recommendation method comprising a.
The method of claim 6,
The extracting step,
Selecting a reference product from among the first products to compare product preferences with the other users;
Extracting the user's first product preference for the reference product
Extracting second product preferences of the other users for the reference product
The first product preference and the second product preference are compared, and based on the comparison result, users similar to the first product preference among the other users are extracted as users having a product preference similar to the product preference of the user. Steps to
Online product recommendation method comprising a.
The method of claim 6,
Predicting the product preference,
Extracting a product preference of the user with respect to a product selected by the user's action on the product from among the first products;
Calculating a similarity between the user and the extracted users based on product preferences of the user for the selected product and product preferences of the extracted users for the selected product; And
Predicting product preference of the user with respect to the third products based on the similarity
Online product recommendation method comprising a.
The method of claim 6,
Providing the product recommendation list,
Calculating a ranking value representing a recommended ranking for products in the product list based on the predicted product preference and the product reliability; And
Providing the product recommendation list according to the ranking value
Online product recommendation method comprising a.
A preference determiner for determining a product preference of the user with respect to the first products based on the user's online behavior with respect to the first products;
A reliability determiner for determining product reliability for the second products based on online behavior of other users with respect to the second products; And
Including; a recommender for providing a product recommendation list to the user based on the product preference and the product reliability,
The user's online behavior,
Including a product evaluation score of the user, the number of product clicks, whether a product is steamed for future purchase of the first products, and a time spent on the product page,
The online behavior of the other users,
Including the number of product mentions by the other users, product evaluation scores, product clicks, product wishes for future purchases of the second products, time spent on product page, and number of times stayed on product page
Online product recommendation device.
The method of claim 10,
The preference determiner,
Extracting an action value for each online action performed on each of the first products,
Calculate an action weight for each online action based on the action value,
Calculate the user's current product preference for the first products based on the action value and the action weight,
Calculating the product preference based on the current product preference and the user's previous product preference for the first products
Online product recommendation device.
The method of claim 11,
The preference determiner,
Normalize the behavior value,
Using the normalized action value, the percentage of each online action is calculated,
Calculating the action weight based on the performance ratio and the number of the first products evaluated by the user
Online product recommendation device.
The method of claim 11,
The preference determiner,
Calculate time weights for the first products based on the current product preference and the previous product preference,
Calculating the product preference by reflecting the time weight on the current product preference
Online product recommendation device.
The method of claim 10,
The reliability determiner,
Calculate the degree of interest of the other users with respect to the second products based on the product evaluation score of the other users, the number of wishes, the time spent on the product page, and the number of times they stayed on the product page,
Calculating the product reliability based on the number of mentions of the product and the degree of interest
Online product recommendation device.
The method of claim 10,
The recommender,
Comparing product preferences for the first products of the other users and product preferences for the first products of the user to extract users with product preferences similar to the product preferences of the user from among the other users,
Predicting the user's product preference with respect to the third products based on the extracted user's product preference with respect to the third products,
Sorting the third products according to the predicted product preference to generate a product list,
Providing the product recommendation list based on the product list and the product reliability
Online product recommendation device.
The method of claim 15,
The recommender,
Selecting a reference product from among the first products to compare product preferences with the other users,
Extracting the user's first product preference for the reference product,
Extracting second product preferences of the other users for the reference product,
The first product preference and the second product preference are compared, and based on the comparison result, users similar to the first product preference among the other users are extracted as users having a product preference similar to the product preference of the user. doing
Online product recommendation device.
The method of claim 15,
The recommender,
Extracting the user's product preference for the product selected by the user's action on the product from among the first products,
Calculate a degree of similarity between the user and the extracted users based on product preferences of the user for the selected product and product preferences of the extracted users for the selected product,
Predicting the user's product preference for the third products based on the similarity
Online product recommendation device.
The method of claim 15,
The recommender,
Calculate a ranking value representing a recommendation ranking for products in the product list based on the predicted product preference and the product reliability,
Providing the product recommendation list according to the ranking value
Online product recommendation device. delete delete

Images