KR20160015858A - The restaurant recommending system and the recommending method thereof - Google Patents

Abstract

An example is disclosed to provide a restaurant recommending system, comprising: a communication unit for receiving a restaurant recommending request from a user terminal of a specific user through a wire or wireless communication network; a header for performing a web crawler and a wrapper-based web crawler linked with a plurality of storage servers and storing the characteristics of a plurality of restraints prepared by a plurality of users; and a controller for receiving a request recommending signal from the communication unit, selecting a candidate having a similar profile to a specific user by receiving the characteristics of the restaurant, and recommending the restaurant having a high similarity after comparing the characteristics of the restaurant prepared by the selected candidate and the profile of a specific user. Accordingly, by analyzing the big data and recommending a restaurant satisfying the self congruity between the image of a restaurant and the image of a user, the satisfactory can be improved.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a restaurant recommanding system

The present invention relates to a method of recommending a restaurant. More particularly, the present invention relates to a system and method for recommending a restaurant that matches a user by analyzing an image of the restaurant from a review of restaurants collected using a big data analysis technique.

The consumer considers various attributes such as restaurant location, food quality, service quality, atmosphere quality, price, past purchasing experience, and brand when making a purchase decision for dining out.

Consumers choose brands based on their own evaluations and beliefs about them.

On the other hand, as the analysis of data that was too large in the past to analyze could be made possible, new values can be found through data generated at near real time.

Today, analytical techniques using these data are becoming reality, and the development of IT for the analysis of big data has been accelerating in recent years. Research is being actively carried out to understand the newly generated data and to extract useful knowledge, to detect and monitor current major issues, and to predict the future.

In addition, big data analysis in the food service industry is being actively pursued through location-based services, and studies are being actively conducted to detect and monitor current major issues by understanding newly generated data and extracting useful knowledge. .

The reason why we are interested in big data in business and academia is that customers' thoughts are in big data.

However, the Big Data Analysis technique, which is an engineering technique for image measurement, has been studied in the field of management and it has been difficult to communicate among the disciplines.

The present invention provides a recommendation system and a recommendation method for recommending a customized restaurant by analyzing big data and grasping a gap between an image of a restaurant and an image of a user.

Embodiments include a communication unit for receiving a restaurant recommendation request from a user terminal of a specific user via a wired or wireless communication network, a web crawler and a wrapper-based web crawler in cooperation with a plurality of storage servers A preprocessing unit for receiving a request for a request from the communication unit, receiving characteristics of the restaurant from the preprocessing unit, and selecting a candidate whose profile is similar to that of a specific user And a controller for comparing the restaurant characteristic created by the selected candidate with the profile of the specific user to recommend the restaurant having a high degree of similarity.

The candidate may be a user with high similarity calculated by the following equation.

는 특정 사용자의 프로파일이고,

는 n명의 다른 사용자의 프로파일이며.

는 각 사용자가 가지고 있는 특징 값에 대한 빈도수이다.At this time,

Is a profile of a particular user,

Is the profile of n other users.

Is the frequency of feature values that each user has.

The preprocessor may perform the web crawler and the wrapper based web crawler to obtain specific data, clean the specific data, perform spam filtering, and analyze the restaurant image from the non-spam data.

The spam filtering can calculate the probability of spam and non-spam classes through the following equation.

는 클래스 확률로서 (클래스 j에서 유일한 단어의 수 / 전체 클래스에서 유일한 단어의 수) 와 같이 계산한다.

는 특정 단어가 특정 클래스에 속할 확률로서 (단어의 빈도수 + 1)*(클래스 j에의 단어 빈도수 + 전체 클래스에서 유일한 단어 수)와 같이 계산한다. At this time,

Is calculated as the class probability (the number of unique words in class j / the number of unique words in all classes).

Is the probability that a particular word belongs to a particular class (frequency of word + 1) * (word frequency in class j + number of unique words in the whole class).

The control server may further include a storage unit for storing each of the restaurant image and the profile of the user.

Meanwhile, the embodiment includes a preprocessing step of storing characteristics of a plurality of restaurants created by a plurality of users by performing a web crawler and a wrapper-based web crawler in association with a plurality of storage servers Receiving a request signal for a restaurant recommendation request of a specific user through a user terminal, selecting a candidate whose profile is similar to that of the specific user, comparing the restaurant characteristic created by the selected candidate with the profile of the specific user And recommending the restaurant having a high degree of similarity.

The present invention can recommend a restaurant that can analyze the big data and meet the self-consistency and self-improvement between the image of the restaurant and the image of the user.

Therefore, accuracy and satisfaction of restaurant recommendations can be improved for individual users.

1 is a conceptual diagram showing an overall system including a restaurant recommendation system of the present invention.
2 is a detailed configuration diagram of a restaurant recommendation system of the present invention.
3 is a flowchart showing the operation of the restaurant recommendation system of the present invention.
4 is a detailed flowchart of the preprocessing step of FIG.
5 is a detailed flowchart showing the web crawler steps of FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between .

Throughout the specification, when an element is referred to as "comprising ", it means that it can include other elements as well, without excluding other elements unless specifically stated otherwise. Also, the terms " part, "" module," and " module ", etc. in the specification mean a unit for processing at least one function or operation and may be implemented by hardware or software or a combination of hardware and software have.

Hereinafter, a restaurant recommendation system will be described in which a restaurant having a high degree of satisfaction is measured by measuring an image difference between a user and a restaurant.

Self image and product image are measured by using 5 attributes (Sincerity, Sophistication, Excitement, Competence, Ruggedness) that measure the characteristics of goods and attractions and compare the difference between them. (Self - conformity), and when the difference is large (self - improvement), the satisfaction of the purchased product increases.

Consumers feel the sense of self-definition by consuming a product or service and communicate it to others. Therefore, the brand is preferred to the brand that reflects and expresses the self concept well. The higher the brand identification, the higher the brand loyalty.

Self congruity theory is the theory that consumption behavior focuses on the symbolic function rather than on the consumption object (Sirgy, Johar, Samli, & Claiborne, 1991; Ekinci & Hosany, 2006). In other words, it can be understood that the consumption of goods or brands is a behavioral theory that emphasizes the aspect of strategic behavior that expresses and strengthens the self image of consumers.

Consumers can satisfy their desire for self-consciousness by using products that have an image similar to the self concept they have. It also meets the desire for self-improvement by using products that have an image similar to the ideal self concept that they want to be. Self-consistency may be related to the product as a whole (Sirgy 1982), or to a specific brand image (Fournier 1998).

Consumers have more favorable attitudes and purchase intentions about brands that they perceive to be consistent with their image. In other words, the correspondence between consumer's perceived self-image and brand image has a significant effect on enhancing consumer's choice, purchase, and after-purchase satisfaction, and generally high preference and loyalty I have.

The most representative types presented in self image studies are real self image and ideal self image. Specifically, the actual self refers to the image of the actual oneself, the ideal self, the image of oneself to want to be, the actual social self, the image of oneself that they actually perceive, and the ideal social self, the image that others want to see.

Therefore, when a brand can prove its self in the process of linking the brand - ego and improve its self, it can be selected and selected by the consumer. In other words, the consumer meets self-desire using a brand that proves self and improves self.

The present invention provides a restaurant recommendation system that can meet consumer expectations based on the theory of self-consistency.

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

FIG. 1 is a conceptual diagram showing an entire system including a restaurant recommendation system of the present invention, and FIG. 2 is a detailed configuration diagram of a restaurant recommendation system of the present invention.

1 and 2, an overall system including a restaurant recommendation system according to an embodiment of the present invention includes a consumer terminal 100, a control server 200, and a storage server 300.

The consumer terminal 100 includes a smartphone, a notebook PC, a tablet PC, or a personal computer, etc., which is connected to the control server 200 and can request the control server 100 to recommend a restaurant.

The consumer terminal 100 interacts with the control server 200 through a wired or wireless Internet, and the wireless Internet may be wifi, Bluetooth, or the like.

The consumer terminal 100 can mount an application for the control server 200. [

The plurality of storage servers 300 may be a storage server for storing data of social media such as various websites, blogs, and tweets.

The control server 200 is a restaurant recommendation system that selects a restaurant optimized for a user after receiving a restaurant recommendation request transmitted from the user terminal 100 via a wired or wireless Internet.

The detailed configuration of the control server 200 is shown in FIG.

2, the control server 200 includes a communication unit 210, a control unit 250, a preprocessing unit 220, and a storage unit 230.

The communication unit 210 includes a communication module for receiving a signal transmitted by wire or wireless, and may include an antenna, a bandpass filter, and the like when receiving a signal wirelessly.

The communication unit 210 receives a request signal transmitted from the user terminal 100 or transmits a result of the request signal.

The communication unit 210 includes a Wi-Fi module (Wi-Fi Module) that can receive a request signal transmitted from an external user terminal 100 using a Wi-Fi communication network when receiving a signal wirelessly, And a serial communication module such as a USB terminal that is connected by a data cable and can receive a request signal by a serial communication method when receiving a signal.

The control unit 250 is a microprocessor. When receiving a restaurant recommendation request from the user terminal 100, the control unit 250 receives restaurant images on various webs preprocessed through the preprocessing unit 220 and performs mutual analysis with the user's image Select the restaurant that has been selected.

The preprocessing unit 220 receives various evaluations of the restaurants in the storage servers 300 connected in accordance with the command of the controller 250 and preprocesses them to generate images of the respective restaurants.

The image of each restaurant thus generated is stored in the storage unit 230.

Hereinafter, the operation of the restaurant recommendation system of the present invention will be described with reference to Figs. 3 to 5. Fig. Hereinafter, the restaurant recommendation system is unified into a control server.

First, when a restaurant recommendation request is received from a specific user via the user terminal 100 to the communication unit 210 (S10), the control server 200 starts a restaurant recommendation operation.

Next, the control unit 250 of the control server 200 obtains information on a specific user and information on various restaurants through the preprocessing unit 220, preprocesses the extracted information, and extracts an image (S20).

More specifically, the preprocessing step will be described with reference to Fig.

First, the web crawler proceeds to obtain information on the web from various storage servers 300 (S21).

The web crawler stage is for collecting data on restaurants from social media such as various websites, blogs, and tweets. Since social media sites, such as Twitter, provide application programming interfaces (APIs) We develop crawlers and collect data.

On the other hand, since a general site such as a blog does not provide an API, a wrapper-based web crawler can be further added to collect data.

Referring to FIG. 5, in the wrapper-based web crawler step, the wrapper-based web crawler refers to the structure information for extracting desired information in the collected web page.

In other words, unlike the general web crawler, the wrapper-based web crawler must directly construct the address of the web page including the information to be extracted and use the structure information of the web page to extract desired data.

First, set the URL to visit.

Such a visit URL setting can be carried out by storing URLs of various blogs directly in the storage unit.

When the wrapper-based web crawler step is started, several URLs of stored URLs are acquired (S211).

At this time, the acquired URL means URL information to be visited, which is stored in the storage unit 230.

Next, the user visits the corresponding URL and performs the page acquisition step of the corresponding content (S213).

When the page having the content is acquired, the wrapping process is performed in which only the data of the desired content is selected using the structure of each site (S215).

If the selected data is stored and it is determined that all the data is selected in the acquired URL, the wrapper base web crawler is terminated and the crawler step is terminated (S219).

In the case of information that does not provide such an API, a wrapper-based web crawler can be separately performed to acquire information of a greater variety of sites.

Next, the pre-treatment clean step of FIG. 4 is performed (S22).

The preprocessing clean step is a process of removing unnecessary characters, symbols, numbers, etc. contained in the acquired web data.

This process is performed in order to reduce errors in data processing in the next step by removing characters expressed as html tag, character / bar, etc. In this step, character of character to be removed is analyzed, and based on the characteristic, Can be defined.

For example, an html tag can be defined as a rule that starts with '<' and ends with '>' as a rule, and can be used to check for string matching if it is a character / If it contains certain characters, you can remove it.

In this way, we analyze the characteristics of characters to be removed in the preprocessing clean step, and define the rules in the program based on the analyzed contents to organize the information.

When the preprocess clean step is finished, the spam filtering step is performed (S23).

The spam filtering step builds a training model and classifies spam documents using training models.

First, unnecessary characters (html tags, symbols, characters / vowels, etc.) are removed from the document after the preprocess clean step. For example, in a document that contains content such as "<b> <h1> both price and feature video quality is satisfactory </ h1> </ b> I'm all satisfied. "

Next, tag the part of speech with the part mark tag. In the above example, you can get the tagging results such as "N_price, N_function, N_video, N_quality, Z_not, N_satisfaction", where N is a noun and Z is an adverb .

Next, one training model is created when the part-of-speech specifies the frequency and class value (eg, spam, non-spam) for each word tagged.

The training model according to the above example can be summarized in the following table.

No. Feature Value Frequency The class value One N_price One Non-spam 2 N_ Function One Non-spam 3 N_ Videos One Non-spam 4 N_ picture quality One Non-spam 5 Z_ everyone One Non-spam 6 N_ Satisfaction One Non-spam 7 V_ good One Non-spam ... ... ...

Next, classification is conducted through the training model.

Remove unnecessary characters (html tags, symbols, letters / vowels, etc.) from the document.

For example "<BR> <H3> I really like it. </ H3> </ H3> </ H3> </ H3> </ H3> I like the image quality and the video quality is good. "

Next, tag the part of speech with the part mark tag. Using the results of the previous step, you can get tagging results like "N_ real, V_ good, N_quality, N_video". Where N is a noun, and V is a verb / adjective.

Finally, the extracted feature values are classified using the following equation.

는 클래스 확률로서 (클래스 j에서 유일한 단어의 수 / 전체 클래스에서 유일한 단어의 수) 와 같이 계산한다.

는 특정 단어가 특정 클래스에 속할 확률로서 (단어의 빈도수 + 1)*(클래스 j에의 단어 빈도수 + 전체 클래스에서 유일한 단어 수)와 같이 계산한다. At this time,

Is calculated as the class probability (the number of unique words in class j / the number of unique words in all classes).

Is the probability that a particular word belongs to a particular class (frequency of word + 1) * (word frequency in class j + number of unique words in the whole class).

For example, if the only number of words in a non-spam class is 18, the frequency of a word is 23, and the only number of words in the spam class is 19 and the frequency of words is 24, the likelihood for non-spam and spam is It can be obtained as follows.

likelihood =

P ("N_Real" | Nonspam) * P ("V_Cost" | Nonspam) * P ("N_Image" (1 + 1) / (24 + 37) * (1 + 1) / (24 + 37) ) * (19/37)

= 5.93407E-07

likelihood =

P ("N_Real" | Spam) * P ("V_ Good" | Spam) * P ("N_Image"

(0 + 1) / (23 + 37) * (0 + 1) / (23 + 37) * 18/37)

= 3.75375E-08

Thus, the non-spam probability is 0.941 and the spam likelihood / (spam likelihood + non-spam likeness) = 0.059.

Therefore, the example sentence is classified as non-spam.

Such a classification can filter spam data and extract only non-spam data.

Next, the image of the restaurant is analyzed using the extracted data.

Using 5 Factor, 15 Facet Name, and 42 Traits of Sincerity, Excitement, Competence, Sophistication, and Ruggedness, It refers to a set of human characteristics associated with a brand and represents the source of benefits that provide intangible value to customers.

Restaurant image analysis extracts data and transforms it into quantitative data and draws meaningful results. It establishes valid criteria, classifies the collected samples into a certain category, and deduces the relationship between the classified categories by statistical methods I do it.

The image measurement items are shown in Table 2 below.

Factor Attribute (FacetName) Traits Expanded metrics (E-Traits) refinement
(sophistication)
Upper-class
(upperclass) The upperclass (upperclass) upperclass Goodlooking goodlooking attractive
(charming) Charming charming, magic, magical, sorcerous, witching, wizard, wizardly Smooth smooth, politic, smooth, suave, bland, fluent, fluid, liquid, legato, placid, quiet, still, tranquil, unruffled Sincerity
(sincerity)
beneficial
(wholesome) Wholesome wholesome The original (original) original honest
(honest) Honest honest, honorable, dependable, reliable, true, good, fair Real real, existent, tangible, actual, genuine, literal, substantial, material, veridical The sincere sincere, earnest, solemn Realistic
(down-to-earth) Down-to-earth down-to-earth, earthy Small-town small-town Family-oriented family-oriented cheerful
(cheerful) Cheerful pollyannaish Friendly friendly, well-disposed, welldisposed, audacious Sentimental sentimental, upbeat, bathetic, drippy, hokey, maudlin, mawkish, kitschy, mushy, schmaltzy, schmalzy, sentimental, sappy, soppy, soupy, slushy Strength
(ruggedness)
rough
(tough) Tough rugged, tough, toughened, sturdy, ruffianly, bad, hard, baffling, elusive, knotty, problematic, problematical Rugged furrowed, rugged, broken Outdoors
(outdoorsy) Outdoor (outdoorsy) outdoorsy Western (western) western, westerly Masculine masculine excitement
(excitement)
brand-new
(up-to-date) Up-to-date up-to-date, uptodate, cutting-edge, with-it Contemporary contemporary, modern-day, present-day, contemporaneous Independent, independent, autonomous, self-governing, sovereign, main active
(spirited) Spirited spirited, bouncing, bouncy, peppy, zippy, game, gamy, gamey, gritty, mettlesome, spunky, enlivened Cool (cool) cool, coolheaded, nerveless, allright, fine, okay, hunky-dory Young young, immature, new, youthful, vernal, unseasoned, untested, untried Imaginary
(imaginative) Imaginative imaginative, inventive Unique unique, alone, unequaled, unequal, unparalleled, singular bold
(daring) Daring daring, audacious, venturesome, venturous Trendy trendy, voguish Exciting exciting ability
(competence)
successful
(successful) Successful successful Confident confident, convinced, positive, surefooted, sure-footed Reliable
(reliable) Reliable reliable, dependable, honest, true, authentic Secure secure, unafraid, untroubled, impregnable, inviolable, strong, unassailable, unattackable, dependable, good, safe intellectual
(intelligent) Intelligent intelligent, well-informed, healthy, levelheaded, level-headed, sound, reasoning, thinking The corporate (corporate) corporate, bodied, corporal, embodied, incarnate, collective, incorporated Technical technical, proficient, technological, mechanical, mechanicallyskillful, expert

That is, if there is an expression for the extended measurement item, the restaurant having one of the five factors is selected.

When an image of each restaurant is selected as described above, it is stored and the preprocessing step is terminated.

When the preprocessing step is finished, a candidate whose information is similar to or different from the user who has requested the recommendation is searched (S30).

When it is necessary to provide recommendation information to a specific user, the similarity of profile information between a specific user and another user is calculated.

The similarity of such profile information can be derived from the following equation.

는 특정 사용자의 프로파일이고,

는 n명의 다른 사용자의 프로파일이며.

는 각 사용자가 가지고 있는 요인 값에 대한 빈도수이다.At this time,

Is a profile of a particular user,

Is the profile of n other users.

Is the frequency of the factor values that each user has.

,

와 같을 때 다음과 같은 유사도 값을 얻을 수 있다.For example, if a profile of a particular user and another user

,

The following similarity value can be obtained.

12 + 22 + 12 + 12 + 12 + 12 + 12 + 12 + 12 + 12 + 12 + 02 * 12 + 02 + 12 + 02 + 12 + 02) = 3 /? 24 = 0.612

When the degree of similarity is calculated, a candidate similar to or different from a specific user is searched.

Next, the feature of the restaurant that the candidate looked at is extracted (S40).

The process of extracting feature values of such restaurants can be extracted by using a character extraction technique such as object name recognition, part-of-speech tagging, bigram, and trigram. In the present invention, however, Which is an example of the feature value extracted through the method of FIG.

Next, the similarity between the profile of the specific user and the candidate restaurant is compared (S50).

The feature value extracted from the document of the candidate restaurant is expressed by the following quantified vector form by calculating the frequency of each feature value.

를 정량화 한 형태로 표현한 것으로

는 요인 값에 대한 빈도수를 의미한다.At this time, Equation (3)

As expressed in quantified form

Means the frequency with respect to the factor value.

와 문서

와의 유사도를 계산한다. 이 때, 코사인 유사도 또는 유클리언 거리 계산 식 등을 통해 유사도를 계산할 수 있다.Profiles of specific users

And documents

Lt; / RTI &gt; In this case, the degree of similarity can be calculated through a cosine similarity degree calculation or an Euclidean distance calculation equation.

When the similarities are calculated, the control server 200 can recommend restaurants to the specific users in descending order of similarity.

At this time, the control server 200 can recommend a restaurant suitable for a specific user by using the 5RRI even when the restaurant is recommended.

To do this, we can analyze the gap of self-congruence by diagnosing the self-consistency of the restaurant and a specific user, and recommend a restaurant that can achieve self-congruity based on the analysis result.

5RRI can be calculated by the following equation.

5RRI (5 Restaurant Recommand Index): restaurant recommendation index, II (Ideal Image): image of the target restaurant AI (Actual Image): image of self, n: And the attribute of the measurement index, respectively.

By analyzing big data like this, it is possible to recommend optimized restaurants by recommending restaurants that have similar or greater self-consistency with users, and can increase satisfaction.

It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (6)

A communication unit for receiving a restaurant recommendation request from a user terminal of a specific user via a wired or wireless communication network,
A preprocessor for storing characteristics of a plurality of restaurants created by a plurality of users by performing a web crawler and a wrapper-based web crawler in association with a plurality of storage servers,
Receiving a request signal from the communication unit, receiving a characteristic of the restaurant from the preprocessing unit, selecting a candidate whose profile is similar to the specific user, comparing the restaurant characteristic created by the selected candidate with the profile of the specific user And a controller for recommending the restaurant having a high degree of similarity
A restaurant recommendation system. The method according to claim 1,
Wherein the candidate is a user having a high degree of similarity calculated by the following equation.

At this time,

Is a profile of a particular user,

Is the profile of n other users.

Is the frequency of feature values that each user has. The method according to claim 1,
The pre-
Performing the web crawler and the wrapper based web crawler to obtain specific data, cleaning the specific data, performing spam filtering, and analyzing the restaurant image from non-spam data. The method of claim 3,
The restaurant recommendation system of claim 1, wherein the probability of the spam and the non-spam class is calculated through the following equation.

At this time,

Is calculated as the class probability (the number of unique words in class j / the number of unique words in all classes).

Is the probability that a particular word belongs to a particular class (frequency of word + 1) * (word frequency in class j + number of unique words in the whole class). 5. The method of claim 4,
Wherein the restaurant recommendation system further comprises a storage for storing each of the restaurant images and the profile of the user. A preprocessing step of storing characteristics of a plurality of restaurants created by a plurality of users by performing a web crawler and a wrapper-based web crawler in association with a plurality of storage servers,
Receiving a request signal for a restaurant recommendation request of a specific user through a user terminal,
Selecting a candidate whose profile is similar to a particular user, and
Comparing the restaurant characteristic created by the selected candidate with the profile of the specific user and recommending the restaurant having a high degree of similarity
&Lt; / RTI &gt;

Images