KR20200133439A - Product recommendation system and method reflecting user purchasing criterion - Google Patents

Abstract

According to one embodiment of the present invention, a system for recommending a product reflecting user purchase criteria includes: a data collection part crawling and collecting product review data and product information regarding a product for which a user has searched, based on purchase criteria about the product selected by the user during a product search from an online shopping mall system; a data preprocessing part conducting morpheme analysis and emotion analysis with respect to the product review data to calculate an emotion score about the purchase criteria by digitizing positive and negative assessments contained in the product review data; and a product recommending part providing products for extracting to be recommended for the user from a product list in accordance with the product information based on the emotional score to provide the products to the user as customized recommendation information. Therefore, the present invention is capable of solving an existing problem that products with a large number of reviews are placed in upper ranks regardless of purchase assessments.

Description

Product recommendation system and method reflecting user purchase criteria {PRODUCT RECOMMENDATION SYSTEM AND METHOD REFLECTING USER PURCHASING CRITERION}

Embodiments of the present invention relate to a product recommendation system and method, and more particularly, a product recommendation system and method that reflects a user purchase criterion capable of providing a product to be recommended to a user by reflecting a purchase criterion selected by the user when purchasing a product. It is about.

User reviews, which allow users to share user opinions on recent products, have been written by various buyers, and are being shared through various channels such as SNS and blogs. These product reviews contain more detailed and reliable information on the preferences of users about the product, and thus can be used usefully in a recommendation system, and many studies are actually being conducted.

In addition, due to the nature of the open web, the amount of product review data created by consumers online has rapidly increased, and various methods have been proposed to naturally analyze the vast amount of opinion information contained in product reviews.

However, the existing product recommendation system is based on information that is not objective or does not consider the user's preference, so the satisfaction level is low. Satisfaction with recommended products can be increased if the criteria that users consider important when purchasing products are considered and reflected.

As a related prior art, there is Korean Patent Publication No. 10-1621938 (name of invention: online shopping information recommendation system using user registration posting information, registration date: 2016.05.11).

An embodiment of the present invention provides a product recommendation system and method that reflects a user purchase criterion capable of providing a product to be recommended to a user by reflecting a purchase criterion selected by a user when purchasing a product.

The problem to be solved by the present invention is not limited to the problem(s) mentioned above, and another problem(s) not mentioned will be clearly understood by those skilled in the art from the following description.

The product recommendation system reflecting the user purchase criteria according to an embodiment of the present invention provides product information and product review data on the product searched by the user based on the purchase criteria for the product selected by the user when searching for a product. A data collection unit for collecting by crawling from an online shopping mall system; A data preprocessing unit that performs morpheme analysis and emotion analysis on the product review data, quantifies positive and negative evaluations contained in the product review data, and calculates an emotional score of the product with respect to the purchase criteria; And a product recommendation unit extracting a product to be recommended to the user from a product list according to the product information based on the emotion score and providing the product as customized recommendation information to the user.

The data preprocessor may select a keyword associated with the purchase criterion, and perform the morpheme analysis on product review data including the selected keyword among the crawled product review data.

The data preprocessing unit extracts a noun representing a characteristic and a purchase criterion, and a verb and an adjective representing an opinion and emotion about the characteristic through the morpheme analysis, and performs the emotion analysis based on the extracted noun, verb and adjective. Can be done.

The data preprocessor counts the number of positive evaluations and negative evaluations among all product reviews for the product based on the result of the sentiment analysis to calculate a positive evaluation ratio and a negative evaluation ratio, and the positive evaluation ratio and the negative evaluation The emotional score including a positive score and a negative score is calculated based on an evaluation ratio, and the sum of the positive score and the negative score is divided by the number of all product reviews to calculate the average value of the emotional score, and the product The recommendation unit may extract the top N products (where N is a natural number) having a high average value of the emotion score as products to be recommended to the user, sort the extracted products in ascending order, and provide customized recommendation information to the user. .

When a plurality of the purchase criteria are selected by the user, the data preprocessing unit assigns a weight for each purchase criteria according to a preset priority when the plurality of selections are selected, and reflects the weight given for each purchase criteria Emotional score can be calculated.

When crawling the product review data, the data preprocessor may divide and read the product review data in sentence units by one sentence based on a period to be useful for later morpheme analysis, and store it in the form of a text file.

The data collection unit stores the collected data in a database (DB), wherein the product information including the URL, product name, and ID of the product is stored in a first table of the database, and the product review data is stored in the database. It is stored in a second table, and product review data including the purchase criteria and a related word related to the purchase criteria may be stored in a third table of the database.

The data preprocessor can perform noise filtering on the product review data to determine affirmative or negative in a sentence of special characters and meaningless stopwords included in the product review data, and the product review data. A sentence that does not contain an adjective may be determined as noise and excluded from the subject of the morpheme analysis.

In the product recommendation method reflecting the user purchase criteria according to an embodiment of the present invention, based on the purchase criteria for the product selected by the user when the data collection unit of the product recommendation system searches for a product, the product searched by the user is Collecting related product information and product review data by crawling from an online shopping mall system; The data preprocessing unit of the product recommendation system performs morpheme analysis and emotion analysis on the product review data to quantify the positive and negative evaluations contained in the product review data, and the emotional score of the product against the purchase criteria Calculating a; And extracting, by the product recommendation unit of the product recommendation system, a product to be recommended to the user from a product list according to the product information based on the emotion score, and providing the product as customized recommendation information to the user.

The performing of the morpheme analysis may include selecting a keyword associated with the purchase criterion; And performing the morpheme analysis on product review data including the selected keyword among the crawled product review data.

When a plurality of purchase criteria are selected by the user, calculating the emotional score of the product may include assigning a weight for each purchase criterion according to a preset priority when the plurality of selections are made; And calculating an emotional score of the product by reflecting the weight assigned to each purchase criterion.

In the product recommendation method reflecting the user purchase criteria according to an embodiment of the present invention, the data preprocessor performs noise filtering on the product review data, so that special characters and meaningless stop words included in the product review data are performed. And determining a sentence in which an adjective for determining positive or negative in the sentence of the product review data is not included as noise and excluding it from the subject of the morpheme analysis.

Details of other embodiments are included in the detailed description and accompanying drawings.

According to an embodiment of the present invention, when purchasing a product, a product to be recommended to a user may be provided by reflecting a purchase criterion selected by the user.

According to an embodiment of the present invention, the number of product reviews and the number of product reviews by using the average value of the emotional score calculated by analyzing product review data morpheme and emotion analysis are used to prevent products with many product reviews from being ranked in the upper ranks. The same score is calculated regardless, and through this, it is possible to solve the existing problem in which products with a large number of product reviews are ranked at the top regardless of purchase evaluation.

According to an embodiment of the present invention, a product optimized for a user's purchase criterion may be ranked at a higher rank by varying the weight for each purchase criterion.

1 is a diagram illustrating a network configuration of a product recommendation system reflecting user purchase criteria according to an embodiment of the present invention.
FIG. 2 is a block diagram illustrating a detailed configuration of the product recommendation system of FIG. 1.
3 is a block diagram illustrating a detailed configuration of the database DB of FIG. 2.
4 and 5 are flowcharts illustrating a product recommendation method reflecting a user purchase criterion according to an embodiment of the present invention.
6 is a diagram showing a morpheme tag provided by the Twitter class and a description thereof.
7 is a case in which “price” is selected as the first priority and “delivery” is selected as the second priority, and is an example of morphological analysis of user review data including keywords related to purchase criteria, sentence by sentence.
8 is a diagram showing the top 10 places as a result of summing positive/negative scores for each purchase criterion.
9 is a diagram showing the weight of the purchase criteria of the top 15 products ranked based on emotion (positive/negative) scores.
10 is a diagram showing the proportion of purchase criteria of the top 15 products.

Advantages and/or features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, only these embodiments make the disclosure of the present invention complete, and common knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to those who have, and the invention is only defined by the scope of the claims. The same reference numerals refer to the same elements throughout the specification.

In addition, a preferred embodiment of the present invention to be implemented below is already provided in each system functional configuration in order to efficiently describe the technical components constituting the present invention, or system functions commonly provided in the technical field to which the present invention belongs. The configuration will be omitted as much as possible, and a functional configuration that should be additionally provided for the present invention will be mainly described. If a person of ordinary skill in the art to which the present invention belongs will be able to easily understand the functions of components previously used among functional configurations that are not shown below and are omitted, the configuration omitted as described above. The relationship between the elements and the components added for the present invention will also be clearly understood.

In addition, in the following description, "transmission", "communication", "transmission", "receive" of a signal or information, and other terms with similar meanings refer to direct transmission of signals or information from one component to another. Not only that, but it includes things that are passed through other components. In particular, "transmitting" or "transmitting" a signal or information to a component indicates the final destination of the signal or information and does not imply a direct destination. The same is true for "reception" of signals or information.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a network configuration of a product recommendation system reflecting a user purchase criterion according to an embodiment of the present invention, and FIG. 2 is a block diagram illustrating a detailed configuration of the product recommendation system of FIG. 1; 3 is a block diagram illustrating a detailed configuration of the database DB of FIG. 2.

1 to 3, the product recommendation system 100 reflecting the user purchase criteria according to an embodiment of the present invention includes a data collection unit 210, a data preprocessor 220, a product recommendation unit 230, A database (DB) 240 and a control unit 250 may be included.

The data collection unit 210 may collect product information and product review data about a product searched by a user by crawling from the online shopping mall system 102. In this case, the data collection unit 210 may collect the product information and product review data based on a purchase criterion for a product selected by a user when searching for a product.

Specifically, the user may access the online shopping mall system 102 through the user terminal 101 and search for a product to be purchased. In this case, the user may select one purchase criterion for the searched product, or alternatively, may select multiple purchase criteria. In this way, when a purchase criterion for the product is selected by the user, the data collection unit 210 uses the purchase criterion for the product to link with the online shopping mall system 102 to obtain a product related to the searched product. Information and product review data may be collected by crawling from the online shopping mall system 102.

Here, the purchase criterion is a criterion prioritized when the user selects a product, and may include durability, design, cost performance, etc. of the product. For example, when the user selects the cost performance as a purchase criterion when searching for the product, the data collection unit 210 selects a review including a sentence related to a price among reviews related to the product, and the selected Reviews can be collected as the product review data.

In addition, the purchasing criteria may further include an influencer. Here, influencer is a collective term for “power bloggers” who operate blogs with great influence on portal sites, social network service (SNS) users with hundreds of thousands of followers, and single-person broadcasters. Influencers can be divided into creator-type influencers that produce content and model-type influencers in the fashion and beauty field. Meanwhile, a marketing method using a person exercising influence is influencer marketing, and the product recommendation system 100 reflecting user purchase criteria according to an embodiment of the present invention can also be applied to influencer marketing.

For example, when the user selects an influencer as a purchase criterion when searching for the product, the data collection unit 210 may select a product related to the influencer among the products searched by the user, that is, a celebrity, a celebrity, or YouTube. Products used or recommended by influencers such as stars may be selected, and reviews related to the selected products may be collected as product review data. In this case, the data collection unit 210 may select a product related to the influencer from among the products searched by the user through data mining.

In this way, when searching for the product, the user may directly select a purchase criterion for the product, but in another embodiment, when searching for the product, the recommended purchase criterion may be applied automatically without selecting the purchase criterion for the product. have. For example, the data collection unit 210 calculates the number of user selections for each purchase criterion by analyzing the purchase criteria for a certain period (eg, one year), and based on the calculated number of user selections for each purchase criterion. Thus, by assigning a priority to each purchase criterion, the purchase criterion of the highest priority may be recommended to the user to aid in the user's selection, or the purchase criterion of the highest priority may be directly applied as a purchase criterion selected by the user.

For another example, the data collection unit 210 calculates the number of user selections for each purchase standard by analyzing the purchase criteria for a certain period (eg, one year), and calculates the number of user selections for each purchase criteria. Based on the weighting of each purchase criterion, the purchase criterion with the highest weight is recommended to the user, but if the weight of the purchase criterion (A) with the highest weight and the weight of the purchase criterion (B) other than the set value is less than the set value, the purchase criterion By recommending both (A) and purchase criteria (B) to the user, the purchase criteria preference list (A, B) may be provided to the user to directly select the purchase criteria preferred by other users.

For another example, the data collection unit 210 calculates the number of user selections for each purchase criterion according to gender and age-specific conditions by analyzing the purchase criteria for a certain period (eg, one year) based on the conditions for each gender and age, and , Based on the calculated number of user selections for each purchase criterion, a weight may be assigned to each purchase criterion.

The data collection unit 210 may store the collected data (product information, product review data) in a database (DB) 240. To this end, the database 240 may include a first table 310, a second data 320, and a third data 330 as shown in FIG. 3.

In other words, the data collection unit 210 stores product information including the URL, product name, and ID of the product in the first table 310 of the database 240, and the product review data is the database 240 ) Can be stored in the second table 320. Further, the data collection unit 210 may store product review data including the purchase criteria and a related word related to the purchase criteria in the third table 330 of the database 240.

When crawling the product review data, the data preprocessing unit 220 may divide and read the product review data in sentence units by one sentence based on a period to be useful for later morpheme analysis, and store it in a text file format.

The data preprocessor 220 performs noise filtering on the product review data, determines special characters and meaningless stop words included in the product review data as noise, and excludes them from the subject of the morpheme analysis. I can. In addition, the data preprocessing unit 220 may determine a sentence in which the adjective for determining positive or negative is not included in the sentence of the product review data as noise and exclude it from the target of the morpheme analysis.

To this end, the data preprocessor 220 may perform noise filtering by removing special characters and meaningless stop words included in the product review data. In addition, the data preprocessor 220 may perform noise filtering by removing a sentence in which an adjective for determining positive or negative is not included in the sentence of the product review data.

The data preprocessing unit 220 performs morpheme analysis and emotion analysis on the product review data to quantify the positive and negative evaluations contained in the product review data, and the emotional score of the product with respect to the purchase criterion Can be calculated.

Specifically, the data preprocessor 220 may select a keyword related to the purchase criterion and perform morpheme analysis on product review data including the selected keyword among the crawled product review data. The data preprocessing unit 220 extracts a noun representing a characteristic and a purchase criterion, and a verb and an adjective representing an opinion and feeling about the characteristic through the morpheme analysis, and based on the extracted noun, verb and adjective. Emotion analysis may be performed on the product review data.

The data preprocessor 220 may calculate a positive evaluation ratio and a negative evaluation ratio by counting the number of positive evaluations and negative evaluations among all product reviews for the product based on the result of the emotion analysis. The data preprocessor 220 may calculate an emotion score including a positive score corresponding to the positive evaluation and a negative score corresponding to the negative evaluation based on the positive evaluation ratio and the negative evaluation ratio. The data preprocessor 220 may calculate an average value of the emotion score by dividing the sum of the positive score and the negative score by the total number of product reviews.

Specifically, the data preprocessing unit 220 morphemely analyzes the product review data by dividing each product, and based on the morpheme analysis result, the number of positive and negative evaluations among all product reviews for each product. Can be counted. The data preprocessing unit 220 calculates a positive evaluation ratio based on the number of positive evaluations relative to the number of total product reviews for each product, and negative based on the number of negative evaluations relative to the number of all product reviews for each product. You can calculate the evaluation rate. The data preprocessing unit 220 calculates a positive score corresponding to the positive evaluation based on the calculated positive evaluation ratio, and calculates a negative score corresponding to the negative evaluation based on the calculated negative evaluation ratio. Emotional scores can be derived. The data preprocessor 220 may calculate the average value of the emotional score for the product by performing a division operation in which the emotion score, that is, a sum of the positive score and the negative score, is divided by the number of all product reviews.

As described above, in an embodiment of the present invention, the same score is calculated regardless of the number of product reviews by using the average value of the sentiment score for the product so that products with many product reviews are not ranked in the top ranking. It is possible to solve the existing problem that products with a large number of reviews are ranked at the top regardless of purchase evaluation.

Meanwhile, the user may select a plurality of purchase criteria when searching for the product. In this case, the user may preset priorities for each of the plurality of purchase criteria. When a plurality of purchase criteria are selected as described above, the data preprocessor 220 assigns weights for each purchase criteria according to a preset priority, and reflects the weights assigned for each purchase criteria to calculate the emotional score of the product. It can also be calculated.

The product recommendation unit 230 extracts a product to be recommended to the user from a product list according to the product information, based on the emotion score calculated by the data preprocessor 220, and selects the extracted recommended product. It can be provided as customized recommendation information to the user.

At this time, the product recommendation unit 230 extracts the highest N (the N is a natural number), for example, 10 products to be recommended to the user with a high average value of the emotion score, and the extracted products in ascending order. It can be sorted and provided as customized recommendation information to the user.

The database 240 may store data collected by the data collection unit 210. In particular, the database 240 may distribute and store the collected data in the first to third tables 310, 320, and 330 as described above.

Specifically, product information including a URL, a product name, and an ID of the product may be stored in the first table 310, and the product review data may be stored in the second table 320. In addition, the third table 330 may store product review data including a purchase criterion selected by the user when searching for a product and a related word related to the purchase criterion.

The control unit 250 is a product recommendation system 100 reflecting the user purchase criteria according to an embodiment of the present invention, that is, the data collection unit 210, the data preprocessing unit 220, and the product recommendation unit 230 , It is possible to control overall operations of the database 240 and the like.

The apparatus described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It can be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

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.

4 and 5 are flowcharts illustrating a product recommendation method reflecting a user purchase criterion according to an embodiment of the present invention.

The product recommendation method described here is only one embodiment of the present invention, and various steps may be added as needed, and the following steps may also be implemented by changing the order, so that the present invention will be described below. It is not limited to each step and the order to be described.

First, referring to FIGS. 1, 2 and 4, in step 410, the user may access the online shopping mall system 102 through the user terminal 101 and search for a product to be purchased.

Next, in step 420, the user may select a purchase criterion for the product when searching for the product. Here, the purchase criterion is a criterion prioritized when the user selects a product, and may include durability, design, cost performance, etc. of the product. In addition, the user may select a large number of purchasing criteria for the product, and in this case, priority may be set in advance for each purchasing criteria.

Next, in step 430, the data collection unit 210 of the product recommendation system 100 may collect product information and product review data about the product searched by the user.

Next, in step 440, the data collection unit 210 of the product recommendation system 100 may store the collected data (product information, product review data, etc.) in the database 240.

Next, in step 450, the data preprocessing unit 220 of the product recommendation system 100 performs morpheme analysis and emotion analysis on the product review data, and calculates the emotional score of the product for the selected purchase criterion. Can be calculated. This will be described in detail with reference to FIG. 5 as follows.

That is, in step 510, the data preprocessor 220 may extract nouns representing features and purchasing criteria, and verbs and adjectives representing opinions and feelings about the features through morpheme analysis. Thereafter, in step 520, the data preprocessor 220 may perform an emotion analysis on the product review data based on the extracted nouns, verbs and adjectives. Thereafter, in step 530, the data preprocessing unit 220 counts the number of positive evaluations and negative evaluations among all product reviews for the product, based on the result of the emotion analysis, and the positive evaluation ratio and the negative evaluation ratio. Can be calculated. Thereafter, in step 540, the data preprocessor 220 generates an emotion score including a positive score corresponding to the positive evaluation and a negative score corresponding to the negative evaluation based on the positive evaluation ratio and the negative evaluation ratio. Can be calculated. Thereafter, in step 550, the data preprocessor 220 may calculate an average value of the emotion score by dividing the sum of the positive score and the negative score by the number of all product reviews.

Referring back to FIG. 4, in step 460, the product recommendation unit 230 of the product recommendation system 100 recommends the product to the user from the product list based on the emotional score calculated by the data preprocessing unit 220. You can extract products to do. In this case, the product recommendation unit 230 may extract the top N products (where N is a natural number) having a high average value of the evaluation score as products to be recommended to the user.

Next, in step 470, the product recommendation unit 230 of the product recommendation system 100 sorts the extracted products in ascending order and provides them as customized recommendation information to the user through the user terminal 101. have.

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 CDROMs and DVDs, and magnetic-optical media such as floptical disks. And hardware devices 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.

Hereinafter, an example of implementing a product recommendation system reflecting user purchase criteria according to an embodiment of the present invention will be described in detail.

System implementation

The computer used in this experiment was a Windows operating system based CPU (Intel i5-4690), RAM-8G, and Pycharm was used for implementation. For the system implementation, a case where the user selected price and delivery as the purchase criteria was cited as an example. That is, a case where a user wants to purchase a product with a lower price and faster delivery than other purchasing standards such as design or brand recognition is cited as an example.

1. Data Crawling

For the experiment, Python is used to select'electric heaters' of seasonal home appliances among the categories of domestic shopping malls, and collect information on electric heater products and reviews written by users about the products. Because the quality of product reviews influences this experiment, it was decided as a home appliance with a lot of relatively honest user reviews, and it was collected from Internet sites that recorded more than 7-8 trillion transactions in Korea. As the transaction amount is large, consumers use it a lot, and as much as they use it, there will be as much review data. The product name, product URL, and product review of the product item'Electric Heater', which has the most reviews in the'Seasonal Appliances' category, are collected using BeautifulSoup web crawler. Only products with more than 20 reviews are selected and extracted so that they can be analyzed data.

For product reviews provided by the online shopping mall system to be crawled, the list of all reviews can be viewed only when a certain number of reviews are written, such as clicking the'More' button, and an action on the web browser to turn the page. Therefore, Selenium, a web page testing tool that can be automatically executed when crawling these actions, is used. This tool is initially used to test the web by adjusting the browser, and it has functions to crawl the tag name in the web page or the text in the HTML tag, including JavaScript, so it is possible to collect data easily and is widely used when crawling. It is also a used tool.

The Selenium package runs a web browser through the URL of the product in a Python environment, and collects the product name, product classification, and reviews using the HTML tag name. When crawling product information and product reviews, only products with more than 20 product reviews are stored in the DB.

2. Pre-processing

As above, when crawling product information and product reviews, product reviews, which are qualitative information, must be quantified as quantitative information, so there must be an appropriate level of product reviews. This is because if the weight is calculated with product reviews below a certain level, it is not appropriate when evaluating the product.

Therefore, in this embodiment, only products having more than 20 product reviews are stored in the DB. In order to calculate the score of the product feature, it is necessary to obtain the score of the sentence to which each feature belongs. To do that, it is necessary to separate the sentences. It is common for each review to consist of as few as 1 and as many as 10 sentences. Therefore, when crawling product reviews, each sentence was read based on the period and saved as a text file to facilitate later morpheme analysis. It also performs preprocessing to remove special characters or meaningless stopwords such as'ㅋㅋㅋㅋ'.

In the present embodiment, not only a noun expressing a purchasing criterion but also an adjective describing a purchasing criterion keyword such as'price' is considered. For example, since a user wants a product with a low price, adjectives such as'cheap-' or'cheap-' along with the keyword'price' are added to the keyword and extracted from the user review data. By considering adjectives together with nouns, it is possible to more accurately extract products that meet the purchasing criteria desired by the user, and to filter unnecessary data by considering only information necessary for analysis.

3. Morphological analysis

Install'KoNLP', a package that handles Hangul in Python programs, and execute it based on the Twitter class. The Twitter class is an open source Korean tokenizer written in Scala created by Twitter, and it outputs results more concise and easy to understand. 6 is a diagram showing a morpheme tag provided by the Twitter class and a description thereof. The Twitter class has about 20 tags including a Twitter Hashtag, a Twitter ID (ScreenName), and an email address (Email) not included in FIG. 6.

The text file saved in the previous step is read and saved as a list to perform morpheme analysis one sentence by one. In this process, sentences that do not contain adjectives that can judge positive or negative among the sentences of the review are judged as noise and removed from the analysis target.

7 is a case where'price' is selected as the first priority and'delivery' is selected as the second priority, and is an example of morphological analysis of user review data including keywords related to purchase criteria one sentence by one.

4. Emotional (positive/negative) score assigned

For product recommendation, among the product reviews extracted previously, it is determined whether there are many product reviews evaluated positively or as many negative product reviews as the purchase criteria selected by the user. Sentiment (positive/negative) scores are scored with'adjective' or'verb' that evaluates the purchasing criteria in sentences containing the purchasing criteria of the 1st and 2nd, respectively. For example, if there is a review sentence that says "Delivery is really fast", it gives a +1 point because it expresses a positive expression of the purchase criterion of'delivery' as'fast.' In this way, positive evaluation is +1 point and negative evaluation is -1 point.

When a review that includes the keyword'price' among all reviews is called PR={pr ₁ , pr ₂ , ..., pr _n }, and the review with a positive/negative score is PS (PriceScore), The final keyword review for the purchase standard'price' can be obtained with KR ₁ =PS/RP. Likewise, in the case of the second keyword,'Shipping', the review containing the keyword is DR={dr ₁ , dr ₂ , ..., dr _n }, and the positive/negative score of the review is DS If (DeliveryScore), the final review score of the keyword'delivery' can be calculated as KR ₂ =DS/DR.

However, since neutral opinions were difficult to determine, they were judged as noise and excluded. Product reviews that had positive overall product reviews but did not have good evaluation of the sorting keyword'price' or'shipping' were also included in the exclusion list.

The weight of each purchase criterion can be obtained by multiplying the score of the positive and negative scores by the weight of each purchase criterion and expressing it as a percentage. It can be seen that the higher this ratio, the higher the probability of being ranked in the top. In the same way, the second order of purchase criteria is calculated, and the sum of the two scores becomes the final score, and this score is used to rank. This can be expressed in Equation 1 below.

[Equation 1]

As a result, the product that received the best evaluation among the entire product list is recommended to the user. 8 is a diagram showing the top 10 places as a result of summing positive/negative scores for each purchase criterion.

The product recommendation system proposed in this embodiment collects data using the Beautifulsoup web crawler, and provides recommendation information suitable to users through morpheme analysis and sentiment analysis of product review data in a Python environment. In addition, in order to derive user-tailored recommendation information, purchasing criteria that influence product purchase decisions are directly selected and reflected in the algorithm to differentiate from the existing recommendation system.

The weight can be calculated by dividing the sentiment score calculated earlier by the number of total product reviews. This allows us to see the share of the purchase criteria in the product. And it can be seen that the weight has a great influence on the creation of the product list.

9 is a diagram showing the weight of the purchase criteria of the top 15 products ranked based on emotion (positive/negative) scores. There are products that rank lower than that, even though they have a higher proportion than those in the top 10, such as the 11th product. This shows that the ranking is determined according to the weights of the first and second priority.

10 is a diagram showing the proportion of purchase criteria of the top 15 products. In Figure 9, when comparing the case of the products ranked 11th despite the high ratio and the products ranked top 5 despite the low ratio, the emotional score of'price', which is the first priority than the emotional score of'delivery', which was the second priority It can be seen that the ratio is much higher. Since the weight of the 2nd priority was reflected as half of the weight of the 1st, the ratio of the total review was larger, but the result was not ranked in the top.

In this way, it can be seen that the recommendation order changes according to the priority of the purchase criteria selected by the user. Furthermore, this result may mean that it is possible to recommend customized products for each individual user by reflecting the user's preference and purchasing tendency. Therefore, by varying the weight for each purchase criterion, a product optimized for the user's purchase criterion can be ranked at the top.

As described above, although the embodiments have been described by the limited embodiments and drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. 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 equivalents to the claims also fall within the scope of the following claims.

100: product recommendation system
101: user terminal
102: online shopping mall system
210: data collection unit
220: product evaluation unit
230: product recommendation unit
240: DB
250: control unit
310: first table
320: second table
330: third table

Claims (12)

A data collection unit that crawls and collects product information and product review data about the product searched by the user from the online shopping mall system based on a purchase criterion for the product selected by the user when searching for a product;
A data preprocessing unit that performs morpheme analysis and emotion analysis on the product review data, quantifies positive and negative evaluations contained in the product review data, and calculates an emotional score of the product with respect to the purchase criteria; And
A product recommendation unit extracting a product to be recommended to the user from a product list according to the product information based on the sentiment score and providing customized recommendation information to the user
Product recommendation system reflecting the user purchase criteria, characterized in that it comprises a.
The method of claim 1,
The data preprocessor
A product recommendation system reflecting user purchase criteria, characterized in that a keyword related to the purchase criteria is selected, and the morpheme analysis is performed on product review data including the selected keyword among the crawled product review data.
The method of claim 1,
The data preprocessor
Through the morpheme analysis, a noun representing a characteristic and a purchase criterion, and a verb and an adjective representing an opinion and emotion on the characteristic are extracted, and the emotion analysis is performed based on the extracted noun, verb and adjective. A product recommendation system that reflects the user purchase criteria.
The method of claim 1,
The data preprocessor
Based on the result of the sentiment analysis, the number of positive evaluations and negative evaluations among all product reviews for the product is counted to calculate a positive evaluation ratio and a negative evaluation ratio, and based on the positive evaluation ratio and the negative evaluation ratio The emotion score including the positive score and the negative score is calculated, and the sum of the positive score and the negative score is divided by the number of all product reviews to calculate an average value of the emotion score,
The product recommendation section
Extracting the top N products (where N is a natural number) having a high average value of the emotional score as products to be recommended to the user, and providing the extracted products as customized recommendation information to the user by sorting the extracted products in ascending order. Product recommendation system reflecting user purchase criteria.
The method of claim 1,
The data preprocessor
When a plurality of purchase criteria are selected by the user, a weight is assigned to each purchase criteria according to a preset priority when the plurality of selections are selected, and the emotional score of the product is calculated by reflecting the weight assigned to each purchase criteria. Product recommendation system reflecting the user purchase criteria, characterized in that.
The method of claim 1,
The data preprocessor
When crawling the product review data, the product recommendation system reflecting the user purchase criteria, characterized in that, for use in later morphological analysis, divided into sentences by sentence units based on periods, read and stored in a text file format.
The method of claim 1,
The data collection unit
The collected data is stored in a database (DB), the product information including the product URL, product name, and ID is stored in a first table of the database, and the product review data is stored in a second table of the database. And storing product review data including the purchase criteria and a related word related to the purchase criteria in a third table of the database.
The method of claim 1,
The data preprocessor
By performing noise filtering on the product review data, special characters and meaningless stop words included in the product review data, and adjectives that can determine positive or negative in the sentence of the product review data are included. A product recommendation system reflecting user purchase criteria, characterized in that the non-existent sentence is determined as noise and is excluded from the morpheme analysis.
In the product recommendation method using a product recommendation system reflecting user purchase criteria,
The data collection unit of the product recommendation system crawls product information and product review data about the product searched by the user from the online shopping mall system, based on the purchase criteria for the product selected by the user when searching for the product. Collecting;
The data preprocessing unit of the product recommendation system performs morpheme analysis and emotion analysis on the product review data to quantify the positive and negative evaluations contained in the product review data, and the emotional score of the product against the purchase criteria Calculating a; And
Extracting, by a product recommendation unit of the product recommendation system, a product to be recommended to the user from a product list according to the product information based on the emotion score, and providing the product as customized recommendation information to the user
Product recommendation method reflecting the user purchase criteria, characterized in that it comprises a.
The method of claim 9,
The step of performing the morpheme analysis
Selecting a keyword associated with the purchase criterion; And
Performing the morpheme analysis on product review data including the selected keyword among the crawled product review data
Product recommendation method reflecting the user purchase criteria, characterized in that it comprises a.
The method of claim 9,
When a plurality of the purchase criteria are selected by the user,
The step of calculating the emotional score of the product
Assigning a weight for each purchase criterion according to a preset priority when the plurality of selections are made; And
Calculating the emotional score of the product by reflecting the weight assigned to each purchase criterion
Product recommendation method reflecting the user purchase criteria, characterized in that it comprises a.
The method of claim 9,
The data preprocessing unit performs noise filtering on the product review data, so that special characters and meaningless stop words included in the product review data, and positive or negative in the sentence of the product review data can be determined. Determining a sentence without an adjective as noise and excluding it from the subject of the morpheme analysis
Product recommendation method reflecting the user purchase criteria, characterized in that it further comprises.

Images