KR101218141B1 - Method and system for calculating ranking - Google Patents

Abstract

PURPOSE: A ranking calculation method and a system are provided to use a document without having a hyperlink structure, thereby calculating rankings between objects. CONSTITUTION: A restaurant information collection unit(100) collects information related to restaurants which are searched on the web. The restaurant information collection unit stores the information related to the restaurants in a restaurant database. A review collection unit(110) collects a restaurant review document which is searched on the web. The review collection unit stores the restaurant review document in a review database. An intermediate table generation unit(120) generates an intermediate calculation table and includes an intermediate database which manages the intermediate calculation table. A ranking calculation unit(130) calculates a final ranking value of the restaurants by using the intermediate calculation table. [Reference numerals] (100) Restaurant information collection unit; (102) Restaurant DB; (110) Review collection unit; (112) Review DB; (120) Intermediate table generation unit; (130) Ranking calculation unit; (140) Ranking result storage unit; (142) Ranking DB; (A) Web

Description

Ranking calculation method and system {METHOD AND SYSTEM FOR CALCULATING RANKING}

The present invention relates to a ranking calculation method and system, and more particularly, to a ranking calculation method and system for calculating a ranking using a review document having no hyperlink structure.

Ranking technique is a technique that a search engine calculates to return a document related to a user's query.

The initial ranking technique in foreign countries was determined by reflecting the structure of hyperlinks and the frequency of word appearance in the document for general web documents. After that, the technology was applied.

Domestic ranking technology is a form that uses the initial foreign ranking technology or there is a technology that targets blog documents.

As a foreign technology, the web document ranking technique includes a technique of analyzing a hyperlink included in a document, a method of analyzing the contents of the web document, and determining a rank based on a similarity score between the query and the web document. In the link analysis method, since one web document can give a meaning of recommendation by connecting a hyperlink to another web document, the document which receives a lot of hyperlinks is determined to have received more recommendation and gives a high ranking.

In the link analysis method, a web document is represented using a graph structure composed of nodes and directed edges.

1 is a diagram illustrating an example in which a web document is represented graphically.

Where nodes are web documents and edges are hyperlinks.

At this time, the web document u refers to the web document v, and the web document v is referred to as being referred to by the web document u. The edges are directional depending on whether the web document references or receives them. The edge when the web document u refers to the web document v is called forward, and the edge that web document v refers to by the web document u is called reverse. .

In FIG. 1, the web document v is also referred to by the web document u and refers to the web document w.

InDegree is one of the foreign ranking algorithms, which means that the quality of documents is high when there are many documents that receive a lot of references.

Web document v calculates the ranking of the web document by the number of edges coming in the reverse direction, such as web document u, rather than forward edges such as web document w, that is, the number of other web documents including web document u refer to web document v. do.

PageRank calculates a web document that is referenced from many high quality web documents as a quality web document. The quality of a web document in PageRank assumes a model of randomly visiting and navigating a page called a random surfer.

For example, if page A has three links to pages B, C, and D, B gets 1/3 of A's page rank value, and a surfer who visits page A sees page A and is happy to stop searching. If you are not satisfied with page A, you can visit another page.

If the probability (Damping Factor) is α, page B receives a PageRank of α. When PageRank repeats the value of PageRank exchange between pages through this method, it calculates the final page rank of each page based on the fact that the entire webpage converges a specific PageRank value. Equation 1 shows an equation for calculating PageRank for the i th page p _i .

In Equation 1, d is a damping factor, N is the total number of web pages, M (p _i ) is a set of pages linking p _i pages, and L (p _j ) is the number of outgoing links from p _j pages.

The damping factor is the probability of clicking and moving along a link as described above. Generally, d is 0.85. Analysis of Equation 1 shows that a web page distributes its PageRank evenly to the page to which it is linked. On the page that received the link, the page rank divided by the sum of the damping factors is added to the PageRank value. This process is repeated until the PageRank values converge without any significant change to obtain the final PageRank value.

The HITS ranking algorithm assigns authority and hub scores to a single web page. The score is then used to determine the document quality to calculate the rank. Authority scores as defined by HITS refer to the quality of the content of web documents, and hub scores refer to the degree to which good documents are referenced. A web document has a higher authority score the more it is referred to by a web document with a high hub score. Also, a web document has a higher hub score when more web documents with high authority scores are referred to. Authority scores and hub scores in HITS influence each other. When the authority score is updated, the hub score is newly updated, and when the hub score is newly updated, the authority score is newly updated. When this update is repeated, the authority and hub scores no longer have new values, which are then used to evaluate the quality of the web document.

The SALSA ranking algorithm assigns authority and hub scores to a web document. In SALSA, the authority score calculation process follows the forward edge and the hub score calculation process assumes that the random surfer randomly moves along the reverse edge. At this time, the authority score and the hub score of the web document are determined by the probability that the random surfer stays on the web page.

The Web Page Scoring System (WPSS) algorithm points out that PageRank used by Google's search engine calculates the ranking based on the structure of hyperlinks, not the contents of web documents. In addition, Google supports not only general and horizontal search, but also search of specific domains. In recent years, as the number of vertical portal sites increases, the ranking method that can be applied is considered and suggested. To improve the random surfer model.

As a domestic study, there is a study that proposed a post ranking algorithm for blog environment. This study suggests that the proposed link-based web document ranking algorithm such as PageRank or HITS cannot be applied to post ranking. The reason is that blog posts do not contain hyperlinks. Instead, the algorithm sees the blog's owner as a node concept in PageRank, and uses PageRank's transformed calculation method to post to posts with the notion of scraps, comments, and writebacks that occur in the blog. Determine the ranking for.

Recently, Internet users search for opinions of others via the Internet before visiting restaurants. These comments can be found on sites such as Wingbus, the menu board, and in blog articles. Restaurant search sites such as Wingbus are based on the average value of the user-created star ratings as a standard for restaurant ranking. However, considering only the star rating as a factor of the ranking does not know the exact ranking of the restaurant having the same star rating, there may be distortion in the star rating. For example, one might rate too badly or too badly.

However, the conventional ranking technique is used to determine the ranking of general web documents existing on the Internet. This is because it is generally applied to all kinds of web documents, because it is determined that a document that is referred to through hyperlinks may be of interest to many people. Therefore, it is impossible to apply the conventional ranking technique to the user's review document for the restaurant that does not have a hyperlink structure. This is the same as having to use the number of scraps, the number of comments, the number of posts, etc. to calculate the ranking of blog documents that do not have hyperlinks.

The present invention has been conceived to solve the above problems, and an object thereof is to provide a ranking calculation method and system for calculating a ranking using a review document having no hyperlink structure.

In order to achieve the above object, the ranking calculation system according to the present invention collects information on a plurality of restaurants to be searched on the web, restaurant information collection to manage the database of the information on the plurality of restaurants in a restaurant database And a review collection unit for collecting a plurality of review documents written about restaurants searched on the web, and managing the database of the review documents in a review database, and generating at least one intermediate calculation table used to generate a ranking final result. And an intermediate table generation unit including at least one intermediate database for database-managing the intermediate calculation table, a ranking calculation unit for calculating a final ranking value of each restaurant using at least one intermediate calculation table, and each Having ranked the final ranking value of It includes a ranking result storage unit that is stored in the database and stored in the database.

At this time, the intermediate calculation table is preferably made to reflect the influence score quantified the influence of the users who created the review document for each restaurant.

In addition, the intermediate calculation table may be made so as to reflect the star score that is the numerical evaluation score for each restaurant.

On the other hand, the influence score is preferably made to reflect the recent activity score that quantifies the recent activity for each of the users.

In addition, the influence score is preferably made to reflect a persistence score that quantifies the degree of continuous participation for each of the users.

Further, the influence score is more preferably made to reflect the propagation force score to determine the influence of the user who created the particular review document through the number of review documents at the time after the creation of the particular review document.

Meanwhile, the ranking calculation method according to another aspect of the present invention is made through a ranking calculation system including a restaurant information collecting unit, a review collecting unit, an intermediate table generating unit, a ranking calculating unit, and a ranking result storage unit, and scattered on a web. The information corresponding to a plurality of restaurants are collected in the restaurant information collecting unit of the ranking calculation system, the restaurant information collecting step of being databased in the restaurant database of the restaurant information collecting unit, at least one review document written about a specific restaurant is A review document collection step collected by the review collector and databased in a review database of the review collector, at least one intermediate calculation table used to generate a ranking final result of each restaurant is generated by the intermediate table generator Intermediate table generation step, at least one intermediate system A ranking calculation step in which the final ranking value of each restaurant is calculated by the ranking calculation unit and a final ranking value of each restaurant calculated in the ranking calculation step are databased in the ranking database of the ranking result storage unit by using an acid table. The result storage step is included.

According to the present invention, it is possible to implement a ranking calculation method and system capable of calculating a ranking between ranking calculation targets using a document having no hyperlink structure.

1 is a diagram illustrating an example in which a web document is represented graphically;
2 is a block diagram showing an example of a ranking calculation system according to the present invention;
3 is a flowchart illustrating an example of a ranking calculation method according to the present invention.

In order to explain the ranking calculation method and system according to the present invention, a target of ranking calculation will be defined as a plurality of restaurants (restaurants).

That is, it describes how to find the factors and apply the rankings from review documents scattered on the Internet to calculate the rankings for a plurality of restaurants (restaurants).

2 is a block diagram showing an example of a ranking calculation system according to the present invention.

The ranking calculation system 10 includes a restaurant information collecting unit 100, a review collecting unit 110, an intermediate table generating unit 120, a ranking calculating unit 130, and a ranking result storage unit 140.

The restaurant information collector 100 collects information corresponding to a number of restaurants searched on the web (A). To this end, the restaurant information collecting unit 100 further includes a restaurant database (DB) 102 that manages a database of information on a plurality of restaurants.

In addition, the review collector 110 collects and manages review documents related to restaurants through the web (A). To this end, the review collection unit 110 further includes a review database (DB) 112 for managing a plurality of collected review documents by database.

In addition, the intermediate table generating unit 120 serves to generate an intermediate calculation table used to generate the ranking final result. The information generated as the intermediate calculation table includes, for example, CS information, RFM information, and M information. The intermediate table generation unit 120 uses the CS database 122, the RFM database 124, and the M database to manage them by database. 126 further.

The ranking calculation unit 130 calculates the final ranking value for each restaurant by using the intermediate calculation table information generated by the intermediate table generation unit 120.

The ranking result storage unit 140 stores and manages the final ranking value of each restaurant calculated by the ranking calculation unit 130. For this purpose, the ranking result storage unit 140 further includes a ranking database 142.

3 is a flowchart illustrating an example of a ranking calculation method according to the present invention.

As shown in FIG. 3, the ranking calculation method is performed through a ranking calculation system including a restaurant information collecting unit, a review collecting unit, an intermediate table generating unit, a ranking calculating unit, and a ranking result storage unit, and the restaurant information collecting step (S100). , Collecting document review step S110, intermediate table generation step S120, ranking calculation step S130, and ranking result storage step S140.

In the restaurant information collecting step (S100), information corresponding to a plurality of restaurants scattered on the web is collected in the restaurant information collecting unit of the ranking calculation system, and is databased in the restaurant database of the restaurant information collecting unit.

In the review document collection step (S110), a plurality of review documents written about the restaurant and spread through the web are collected in the review collector, and these review documents are databased in the review database of the review collector.

In the intermediate table generation step (S120), an intermediate calculation table used to generate the ranking final result is generated in the intermediate table generation unit. As the intermediate calculation table, for example, CS information, RFM information, M information, and the like are generated, and these information are databased in the CS database, the RFM database, and the M database, respectively.

In the ranking calculation step (S130), the final ranking value of each restaurant according to the collected individual review documents is calculated in the ranking calculation unit. In this case, a plurality of intermediate calculation tables generated in the intermediate table generation step S120 are used to calculate the final ranking value.

In the ranking result storage step S140, the final ranking value of each restaurant calculated by the ranking calculation unit is managed in the ranking result storage unit. At this time, the final ranking result values are databased in the ranking database.

In the present invention, since the user-created review document, which is the basis for ranking calculation, does not include a hyperlink, existing ranking techniques cannot be applied.

Therefore, the present invention uses Recency, Frequency, Influence, Custormer Satisfaction as ranking factors.

Recency evaluates how recent a user review is.

Frequency evaluates the number of reviews you have written.

Influence determines the influence of a specific user by calculating the number of reviews of a particular restaurant and the number of reviews of a recent restaurant when a user writes a review.

In order to improve the distortion problem in the existing methods that considered only star rating as the ranking factors, we describe how to calculate the ranking for restaurants by applying other factors besides the star rating.

Table 1 is an example of a table for calculating a ranking for a restaurant based on review data.

no shop star title contents name create _ time One Store1 4.5 Title 1 Content 1 people1 2011-05-01 2 Store2 3 Title 2 Content 2 people1 2011-04-30 3 Store1 2 Heading 3 Content 3 people1 2011-04-15 4 Store1 2.5 Title 4 Content 4 people2 2011-04-20 5 Store2 5 Title 5 Content 5 people2 2011-04-15 6 Store1 4 Heading 6 Content 6 people3 2011-04-01 7 Store2 3.5 Heading 7 Content 7 people3 2011-03-30 8 Store1 4 Heading 8 Content 8 people3 2011-02-01 9 Store2 4.5 Title 9 Content 9 people4 2011-02-20 10 Store1 3 Heading 10 Content 10 people4 2011-01-10 11 Store1 2 Heading 11 Content 11 people4 2011-01-01 12 Store1 5 Title 12 Content 12 people5 2011-01-01 13 Store2 5 Title 13 Content 13 people5 2010-12-30 14 Store2 4 Title 14 Content 14 people5 2010-11-30

In order to collect review documents from the review collector, certain rules are required.

The review RT created by the user in Table 1 is defined as Equation 2 to have the following data. Where no is the serial number, shop is the store, star is the star, title is the title of the review, contents is the content of the review, name is the name of the author, and createDate is the date of the review.

On the other hand, the equation used to calculate the ranking of the restaurant in the ranking calculation unit is defined as in Equation 3.

In Equation 3, PI is a measure of the influence of the person who wrote the review as People Influence, CS is a star rating for restaurant j, and reflects the average of several people's star ratings in the final ranking calculation.

That is, as shown in Equation 3, in order to calculate the final ranking of the restaurant, intermediate data such as PI and CS for each restaurant used in the calculation is required.

The intermediate table generator calculates such intermediate data using the techniques described below.

The influence score PI of users who write a review of the store is calculated as shown in Equation 4.

In Equation 4, R is a value for determining recent activity based on RT recently written by P _i , and R (P _i ) is defined as Equation 5 below.

In Equation 5, N is the total number of people. T means the current date at the time of calculating the ranking. The Rank _asc function is an _equal number when sorting in ascending order on a specific set of numbers. The MAX function returns the date of the most recent review written by Pi.

Table 2 is a table showing the results of calculating the R of each people through the equation (5).

P _i N T MAX (t _i ) T-MAX (t _i ) Rank _asc R (P _i ) people1 5 '2011-05-05' '2011-05-01' 4 One 5-1 + 1 = 5 people2 5 '2011-05-05' '2011-04-20' 15 2 5-2 + 1 = 4 people3 5 '2011-05-05' '2011-04-01' 34 3 5-3 + 1 = 3 people4 5 '2011-05-05' '2011-02-20' 74 4 5-4 + 1 = 2 people5 5 '2011-05-05' '2011-01-01' 124 5 5-5 + 1 = 1

In Equation 5, F is a value for determining the degree of continuous participation of the user based on the number of reviews written by P _i as defined in Equation 6.

In Equation 6, N is the total number of people. The COUNT function counts the number of reviews written by P _i at RT. Rank _desc A function is an equal number when sorted in descending order on a specific set of numbers.

Table 3 is a table showing the results of calculating the F of each person through the equation (6).

P _i N COUNT (P _i , RT) Rank _desc F (P _i ) people1 5 3 One 5-1 + 1 = 5 people2 5 2 5 5-5 + 1 = 1 people3 5 3 One 5-1 + 1 = 5 people4 5 3 One 5-1 + 1 = 5 people5 5 3 One 5-1 + 1 = 5

In Equation 4, I is defined as Equation 7 by P _i to write a review about the shop and then determine the influence of Pi based on the amount of reviews written by other people P.

In Equation 8, N is the total number of people. Rank _desc A function is an equal number when sorted in descending order on a specific set of numbers.

In Equation 7, M is the total number of shops that P targets. COUNT (RT _p ) is the total number of reviews in shop s when Pi first wrote a review. COUNT (RT _p ) is the total number of reviews of shop s at the time the ranking is calculated.

Table 4 is a table showing the result of calculating the equation (8).

P _i s COUNT (RT _i ) COUNT (RT _p ) I ₁ (P _i , j) people1 Store1 6 8 0.5 people1 Store2 6 6 0 people1  Sum 0.5 people2 Store1 7 8 0.125 people2 Store2 5 6 0.167 people2  Sum 0.292 people3 Store1 4 8 2 people3 Store2 4 6 0.667 people3  Sum 2.667 people4 Store1 2 8 4.5 people4 Store2 3 6 1.5 people4  Sum 6 people5 Store1 2 8 4.5 people5 Store2 One 6 4.17 people5  Sum 8.67

Table 5 is a table showing the result of calculating the equation (7).

P _i N I ₁ (P _i ) Rank _desc I (P _i ) people1 5 0.50 4 5-4 + 1 = 2 people2 5 0.29 5 5-5 + 1 = 1 people3 5 2.67 3 5-3 + 1 = 3 people4 5 6.00 2 5-2 + 1 = 4 people5 5 8.67 One 5-1 + 1 = 5

In addition, Table 6 is a table showing the result of calculating the final PI value of _Pi through the equation (4).

P _i R (P _i ) F (P _i ) I (P _i ) PI (P _i ) people1 5 5 2 12 people2 4 One One 6 people3 3 5 3 11 people4 2 5 4 11 people5 One 5 5 11

As such, the ranking calculation unit finally calculates the ranking for each restaurant through the intermediate data generated through the intermediate table generation unit.

Next, Table 7 is a table showing the results of calculating the ranking for the restaurant through the equation (3).

j

AVG (CS _j ) AVG (CS _j ) * ω RS Store1 51 3.38 3.38 * 15 = 50.63 101.63 Store2 51 4.17 4.17 * 15 = 62.50 113.50

As described above, the existing ranking technique is based on the structure and form of the web document, but the present invention calculates the ranking based on the influence of each person who writes the review, so the result of the ranking more reasonable than the existing ranking technique Can be derived.

Claims (7)

A restaurant information collection unit which collects information about a plurality of restaurants searched on the web and manages the information about the plurality of restaurants in a restaurant database;
A review collector which collects a plurality of review documents written about restaurants searched on the web, and manages the review documents in a review database;
An intermediate table generation unit including at least one intermediate calculation table used to generate a ranking final result, and including at least one intermediate database for databaseizing and managing the intermediate calculation table;
A ranking calculation unit for calculating a final ranking value of each restaurant using at least one intermediate calculation table;
A ranking calculation system including a ranking result storage unit for storing and managing each of the final ranking value in a ranking database,
The ranking calculation unit may be configured such that a ranking score RANK of a particular restaurant among a plurality of restaurants becomes a sum of the sum of the influence scores of all reviewers for the specific restaurant and AVG (CS), which is an average of the star scores for the specific restaurant. Is calculated using the formula,

,
The influence score PI of each of the reviewers is calculated to be the sum of the recent activity score, the persistence index, and the propagation score of the reviewer,
The intermediate calculation table calculates the influence score, the star score, the activity score, the persistence index and the propagation force score and provides the ranking to the ranking calculation unit.
The recent activity score is a score that quantifies recent activity for each of the users,
The persistence score is a score that quantifies the degree of continuous participation for each of the users,
And the propagation force score is a score that quantifies the influence of the user who created the specific review document through the number of review documents at the time after creation of the specific review document. delete The method of claim 1,
The intermediate calculation table is made to reflect the star scores quantified the evaluation score for each restaurant, wherein the star score is a ranking calculation system that is the average value of the star scores of all reviewers. The method of claim 1,
The recent activity score R (P _i ),
R (P _i ) = {N-Rank _asc (T-Max (t _i ))} + 1
N is the sum of the users, T is the current date at the time of calculating the ranking, and the Rank _asc function is an even number when ascending to a specific set of numbers, and MAX (t _i ) Is a ranking calculation system that is the date of a recent review written by P _i . The method of claim 1,
The persistence score F (P _i ),
F (P _i ) = {N-Rank _desc (COUNT (P _i , RT))} + 1
Where N is the total number of users, COUNT (P _i , RT) is the number of reviews written by P _i in the RT, and Rank _desc is an _equal number when sorting in descending order on a specific set of numbers. Ranking calculation system. The method of claim 1,
The propagation force score I (P _i ),
I (P _i ) = {N-Rank _desc (I1 (P _i ))} + 1
And I1 (P _i ) is

ego,
The N is the sum of the users, the Rank _desc is an equal number when sorted in descending order for a specific set of numbers, the M is the total number of stores targeted by the P, the COUNT (RT _P ) is The total number of reviews of stores when Pi first writes a review, wherein COUNT (RT _p ) is the total number of reviews of stores at the time of calculating the ranking. A ranking calculation method using a ranking calculation system including a restaurant information collecting unit, a review collecting unit, an intermediate table generating unit, a ranking calculation unit, and a ranking result storage unit,
A restaurant information collecting step wherein information corresponding to a plurality of restaurants scattered on the web is collected by the restaurant information collecting unit of the ranking calculation system and is databased in a restaurant database of the restaurant information collecting unit;
At least one review document written about a particular restaurant is collected in the review collector, a review document collection step of being databased in the review database of the review collector,
An intermediate table generation step of generating at least one intermediate calculation table used by the restaurant to generate a final ranking result of each restaurant,
A ranking calculation step of calculating a final ranking value of each restaurant by using the at least one intermediate calculation table in the ranking calculation unit;
A ranking calculation method comprising a ranking result storing step in which a final ranking value of each restaurant calculated in the ranking calculation step is databased in the ranking database of the ranking result storage unit.
The ranking calculation step is such that the ranking score RANK of a particular restaurant among a plurality of restaurants is a sum of the sum of the influence scores of all reviewers for the specific restaurant and AVG (CS), which is an average of the star scores for the specific restaurant. Calculate using the formula below,

,
The influence score PI of each of the reviewers further includes calculating the sum of the reviewer's recent activity score (R), the persistence index (F), and the propagation force score (I).
The intermediate table generating step may further include calculating the influence score, the star score, the activity score, the persistence index, and the propagation force score and providing the ranking to the ranking calculation unit.
In this case, the recent activity score is a score that quantifies recent activity for each of the users, the persistence score is a score that quantifies the degree of continuous participation for each of the users, the propagation power score is after the creation of a specific review document A ranking calculation method that is a score that quantifies the influence of the user who created the specific review document through the number of review documents at the time point.

Images