KR20190005494A - System for recommendating search keywords based on search behavior pattern - Google Patents

Abstract

The present invention provides a system for recommending a search keyword based on a search behavior pattern which can select a search pattern from keywords within a short period. According to an embodiment of the present invention, the system for recommending a search keyword based on a search behavior pattern comprises: a step of calculating vector values of a plurality of words existing in an internet network; a step of using the distance between a vector of a search word inputted from a user terminal for each word and a vector of the words to calculate a distance score; a step of using log data to calculate one or more feature values for the plurality of words; a step of using the one or more feature values for each word to calculate a real-time issue score; and a step of supplying a recommendation search word including at least one word among a plurality of words to the user terminal in accordance with the distance score and real-time issue score. The step of calculating a real-time issue score includes: a step of summing the one or more feature values for each word to calculate a feature score; a step of grouping a plurality of words with an identical search intention among the plurality of words into a single word group to generate at least one word group; and a step of comparing the feature score of the at least one word group (sum of the feature scores of the plurality of words included in the word group) and feature scores of a plurality of words which are not grouped to calculate the real-time issue score.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search word recommendation system based on a search behavior pattern,

The present invention relates to a query term recommendation system based on a search behavior pattern, and more particularly, to a query term recommendation system based on a search behavior pattern configured by extracting a behavior pattern from search data collected from a plurality of users.

 The basic way of recommending a query is to show candidate queries that contain user-entered queries according to their frequency. This does not take into account the context of the user at all, so it is difficult to expect good performance.

 Research called Real-World Context Aware Querying (RECQ) is a research that proposes a method to extend the query considering the context of mobile search. RECQ recommends the ratio of the number of pages when the current user's location name is put together with the query recommendation candidates to the number of corpus documents of all search engines when put into the Google search engine as weights, Suggesting a method of selecting candidates.

 The existing recommendation search service is a service for analyzing a search query input from a user and recommending and providing a highly relevant search word when a user inputs a search word or provides a search result. Conventional suggestion service provided suggestion words using collaborative filtering, but in the case of sparse and large-scale data, there is no intersection with similar patterns, There was a falling problem.

Japanese Patent Application Laid-Open No. 10-2012-0094562, Aug. 27, 2012. open

SUMMARY OF THE INVENTION It is an object of the present invention to provide a query term recommendation system based on a real-time search behavior pattern capable of processing keyword data searched in real time and selecting search patterns from keywords in a short time.

Another problem to be solved by the present invention is to provide a recommendation word provision system capable of providing a recommendation word simply and accurately in the case of a search word having a large amount of data.

The problems to be solved by the present invention are not limited to the above-mentioned problems, and other problems which are not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a search query recommendation system based on a search behavior pattern, comprising: receiving log data including a plurality of search terms input from a plurality of user terminals; Calculating at least one feature value for each search term using data, calculating a score by summing the at least one feature value for each search term, determining a plurality of search terms among the plurality of search terms Generating at least one set of search terms by clustering search terms of a set of search terms into a single set of search terms and generating at least one set of search terms based on the scores of the at least one set of search terms, And ranking the search word rankings .

In some embodiments, the step of estimating the search term rankings may include assigning search term rankings in descending order of scores, wherein in the case of a search term clusters, for a search term with the highest score among a plurality of search terms included in the search term clusters, And the score of the query word cluster can be re-calculated as the sum of the scores of the remaining search words included in the search word cluster.

In some embodiments, the step of generating the at least one set of search terms may group the plurality of search terms into a single set of search terms when the same web document is exposed in a predetermined ratio or more in the search results of the plurality of search words have.

In some embodiments, generating the at least one query word cluster may cluster the plurality of search terms into a single query word cluster if the characters of the plurality of search words are equal to or greater than a predetermined ratio.

Other specific details of the invention are included in the detailed description and drawings.

The search query recommendation system based on the search behavior pattern according to the embodiment of the present invention can accurately analyze the search intention of the user and consider various feature values related to the query amount and the user behavior, Increase the accuracy of issue search queries and minimize reliance on system operators.

A recommendation term is selected according to the distance score between the vector of the term and the word of the word and the real-time issue score of the word, so that it is possible to provide a simple and accurate recommendation word even in the case of a search word having sparse and large data.

The effects of the present invention are not limited to the above-mentioned effects, and other effects not mentioned can be clearly understood by those skilled in the art from the following description.

1 is a block diagram illustrating an environment in which a search query recommendation system based on a search behavior pattern according to an embodiment of the present invention is provided.
2 is a block diagram for explaining a search query recommendation system based on a search behavior pattern according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and the manner of achieving them, will be apparent from and elucidated with reference to the embodiments described hereinafter in conjunction with the accompanying drawings. It should be understood, however, that the invention is not limited to the disclosed embodiments, but may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein, Is provided to fully convey the scope of the present invention to a technician, and the present invention is only defined by the scope of the claims.

The terminology used herein is for the purpose of illustrating embodiments and is not intended to be limiting of the present invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification. The terms " comprises "and / or" comprising "used in the specification do not exclude the presence or addition of one or more other elements in addition to the stated element. Like reference numerals refer to like elements throughout the specification and "and / or" include each and every combination of one or more of the elements mentioned. Although "first "," second "and the like are used to describe various components, it is needless to say that these components are not limited by these terms. These terms are used only to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may be the second component within the technical scope of the present invention.

Unless defined otherwise, all terms (including technical and scientific terms) used herein may be used in a sense that is commonly understood by one of ordinary skill in the art to which this invention belongs. In addition, commonly used predefined terms are not ideally or excessively interpreted unless explicitly defined otherwise.

The terms spatially relative, "below", "beneath", "lower", "above", "upper" And can be used to easily describe a correlation between an element and other elements. Spatially relative terms should be understood in terms of the directions shown in the drawings, including the different directions of components at the time of use or operation. For example, when inverting an element shown in the figures, an element described as "below" or "beneath" of another element may be placed "above" another element . Thus, the exemplary term "below" can include both downward and upward directions. The components can also be oriented in different directions, so that spatially relative terms can be interpreted according to orientation.

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

Referring to FIG. 1, a user terminal 110, a search server 120, and a real-time issue search term selection system 130 are connected to each other through a network. The user terminal 110, the search server 120, and the real-time issue query selection system 130 can exchange data and / or information with each other.

The network may be composed of networks of various sizes such as a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), and the like. The network may be configured as a wired or wireless network.

The user terminal 110 may be a personal computer (PC) such as a desk top, a lap top, or the like. Alternatively, the user terminal 110 may be a portable electronic device such as a smartphone, a personal digital assistant (PDA), a tablet PC, or the like. The user terminal 110 may be a non-illustrated other computing device including a processor, input / output means, and communication means.

The search server 120 provides a search result to the user terminal 110 in response to a search query received from the user terminal 110. The search results may include content such as web documents, images, music, images, files, and the like. The search server 120 can select the content according to a search word (keyword) included in the search query and a search condition. The search server 120 may provide search results listing the contents according to the rank most suitable for the search term. For example, the search server 120 can provide search results in a ranking of high similarity between data such as a search word and a web document. Alternatively, the search server 120 may determine rankings based on freshness of data, quality of data specificity, log of users, and the like.

The search server 120 may be a portal site server that provides various contents services such as a cafe, a mail, a blog, a shopping, a map, a dictionary, a news, have. That is, the search server 120 may be a combination of a search engine and a portal site.

1, the present invention is not limited to this, but a plurality of search servers 120 may be connected to a user terminal 110 and a real-time issue query word sorting system 130 via a network Can be connected.

1, the search server 120 and the real-time issue query word sorting system 130 are separately shown. However, according to the embodiment, the search server 120 may be provided in a form combined with the real-time issue query word sorting system 130 have.

The real-time issue query selection system 130 receives log data from the search server 120 and analyzes the received log data to select real-time issue query words. A real-time issue search query is an issue at the current point of time, which means a query whose query is rapidly increasing. The real-time issue search word selection system 130 transmits real-time issue search word information to the search server 120 so that the search server 120 exposes a real-time issue search word to a web document.

In the embodiment of the present invention, the real-time issue search term selection system 130 minimizes the dependence on the system operator, responds to the attack of the unauthorized users, processes the data in real time, have.

Referring to FIG. 2, a search word selection server 130 of a search query recommendation system based on a search behavior pattern according to an embodiment of the present invention includes a log data receiving unit 131, a filtering unit 132, a feature value calculating unit 133, a score calculating unit 134, a cluster generating unit 135, and a rank calculating unit 136.

The log data receiving unit 131 receives log data from the search server 120. Log data refers to data that records a user's search behavior.

The log data may include a plurality of search terms input from a plurality of user terminals 110. Here, the search term may be defined based on the words input in the search window of the search engine. That is, the search term may include one or more words.

The log data may also include user feedback data for the search results of a plurality of search terms. The user feedback data is used to determine whether or not the user has selected (or clicked) a content such as a web document provided as a search result, a time when the user stays on the search result screen, query, and the like.

The log data may include a plurality of search terms and a time associated with user feedback on search results may be recorded together.

The filtering unit 132 receives log data from the log data receiving unit 131 and filters the received log data.

In some embodiments, the filtering unit 132 may remove the value due to the user's duplicate behavior from the log data received in the filtering process. For example, when the same user repeatedly inputs a query of the same search term for a predetermined unit time, or when the same user repeatedly clicks the same web document or the like provided as a search result, the filtering section 132 May be counted once.

In some other embodiments, the filtering unit 132 may remove values from abusing from the log data received in the filtering process. For example, in the case where only a query is inputted for a predetermined unit time and no click occurs in a web document provided as a search result, when the click count of contents such as a web document provided as a search result is in an abnormal range, The filtering unit 132 may determine that the action is an obtuse and do not count the action.

By filtering the log data in this way, it is possible to respond to an attack by unauthorized users who want to maliciously expose a specific search term to a real-time issue search term. It is to be clearly understood by those skilled in the art that the above-described filtering process is illustrative, and that other filtering methods not exemplarily well-known in the art to filter log data can be used.

The feature value calculation unit 133 receives the filtered log data from the filtering unit 132 and calculates at least one feature value for each search word using the received log data.

The feature value calculating unit 133 may use the statistic of the log data for a predetermined unit time to calculate at least one feature value. The predetermined unit time can be set by the system operator, and can be set to an appropriate time to process data in real time and in a short time. For example, the predetermined unit time may be 15 seconds, but may be modified variously according to the embodiment, but the present invention is not limited thereto.

In an embodiment of the present invention, the feature value calculating unit 133 may use a sliding window corresponding to the unit time to determine the time range of the log data. Specifically, the feature value calculating unit 133 calculates a feature value by a method of determining a time range (for example, a period of 0 second to 15 seconds, a period of 16 to 30 seconds, a period of 31 to 45 seconds, (For example, a period of 0 second to 15 seconds, a period of 1 second to 16 seconds, and a period of 2 seconds to 17 seconds) by using a sliding window method without using the log data ), Real-time issue queries can be processed and updated in real time (every second or every second / minute).

The amount of queries according to the moving average may represent the average amount of queries that occurred during a predetermined time (e.g., 1 minute, 5 minutes, 15 minutes, 30 minutes, 1 hour, 3 hours, etc.). The input frequency and the sustainability of the query of the query can be analyzed according to the query amount according to the moving average. The amount of exposure to the media, such as news, can indicate how much of the news related to the search term is being exposed in the news section of the search result. Real-time community impressions can indicate how many posts (or tweets) related to the search term are being exposed in a real-time community such as Twitter. A click amount of a web document may indicate the number of clicks of contents such as a web document provided as a search result. The activity amount per session and the activity time may indicate the time the user stays on the search result screen. The edited search result screen exposure amount is for considering a case where a search result edited by the search server 120 is provided, such as a search word such as "weather" This is because the search words provided with the edited search results can not be viewed as real-time issue search words. Collection impressions and clicks, such as sites and shortcuts, can indicate how many times they are exposed to the collection, such as the site section and shortcut section of the search results, and the number of clicks. If you click on the shortcut section of a search result to navigate to a site, the search term can not be viewed as a real-time issue search term. However, even if the search result includes a shortcut section, if you click on another section (for example, news, blog, board, cafe article, etc.) without clicking on the shortcut section, the search term will be selected as a real- It will be possible.

In this way, it is possible to increase the accuracy of the real-time issue search word, and consequently, it is necessary for the system operator to directly select the real-time issue search word There is no need to rely on the system operator.

The score calculation unit 134 receives the feature values for each search term from the feature value calculation unit 133, and sums feature values for each search term to calculate a score. The scores calculated for each search term can be used to estimate the search term ranking as described below. For example, the score calculator 134 may add feature values using a linear regression model, but the present invention is not limited thereto.

The cluster generation unit 135 receives the score calculated for each search term from the score calculation unit 134. [

The cluster generating unit 135 groups at least one search word cluster by grouping a plurality of search words having the same search intention among a plurality of search words into one search word cluster. The retrieval intention can indicate a matter (or a purpose, a thought, etc.) that a user wants to acquire through a search query. For example, "My People" and "My Next People" have the same search intent and can be grouped into a single set of search terms.

In some embodiments, the cluster generation unit 135 may group a plurality of search words into a single search word cluster when the same web document is exposed in a predetermined ratio or more in the search results of a plurality of search words. For example, if the same news, blog posts, cafe articles, etc. are exposed in search results for queries of different search terms, the search terms may be grouped.

In some other embodiments, the cluster generator 135 may cluster a plurality of search terms into a single search word cluster if the characters of the plurality of search words are equal to or greater than a predetermined ratio. For example, when the edit distance of a plurality of search words is equal to or less than a predetermined value, the cluster generating unit 135 may group the plurality of search words into one search word cluster. In this case, search terms that are processed with different data due to mistakes in typographical or foreign language marking may be clustered. In addition, the search words in which one or more words are further combined with a specific search word and a specific search word may be grouped together.

The score of the clustered search terms can be expressed as the sum of the scores of a plurality of search terms included in the corresponding search term cluster.

Since the search intention of the user can be accurately analyzed by generating the query word cluster, the accuracy of the real-time issue search word can be improved, and as a result, the system operator does not need to directly select the real-time issue search word. Therefore, Dependency can be minimized. It will be clear to those of ordinary skill in the art that other methods not illustrated may be used to generate the query clusters.

The ranking calculation unit 136 receives the scores calculated from the search word and the keyword cluster information from the cluster generation unit 135 and compares the scores of at least one query word cluster with the scores of the plurality of unclassified search words to calculate a search word ranking do.

The ranking calculation unit 136 can assign the query word rankings in descending order of scores. The ranking calculation unit 136 can assign a ranking of a query word to a search word having the highest score among a plurality of search words included in the search word cluster. The ranking calculation unit 136 may re-score the score of the query word cluster to the sum of the scores of the remaining search terms included in the search word cluster.

Since the score of the first search word Q1 is the highest of 100, the ranking calculation unit 136 can assign the first search word Q1 to the first search word Q1.

Subsequently, the ranking calculation unit 136 determines that the score 99 of the query word cluster including the second and third search terms Q2 and Q3 is higher than the score 77 of the fourth search term Q4, Since the score 66 is higher than the score 33 of the second search word Q2, the second search word Q3 can be assigned the second rank, which is the next rank.

Since the ranking is assigned to the third search word Q3 included in the same search word cluster and the rest of the search words included in the search word group are only the second search word Q2, the second search word Q2 has its own score 33 They will be assigned rankings accordingly. That is, since the score 77 of the fourth search term Q4 is higher than the score 33 of the second search term Q2, the ranking calculation unit 136 assigns the next rank of 3 to the fourth search term Q4, It is possible to assign the fourth rank, which is the next rank, to the search term Q2.

As a result, since the rank of the search word having the highest score in the search word cluster is calculated using the sum of the scores of the search words included in the search word cluster, the rank of the third search word Q3 having a score lower than that of the fourth search word Q4 Can be higher.

The ranking calculation unit 136 may calculate a query word ranking by the above-described method, and may search for a query word up to a predetermined rank (for example, ten) by a real-time issue search word.

The real-time issue search word may be exposed on the web document of the search server 120. The real-time issue search words are listed according to the estimated ranking, and according to the embodiment, the increase of the score may be displayed on the right side thereof.

The steps of a method or algorithm described in connection with the embodiments of the present invention may be embodied directly in hardware, in software modules executed in hardware, or in a combination of both. The software module may be a random access memory (RAM), a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a hard disk, a removable disk, a CD- May reside in any form of computer readable recording medium known in the art to which the invention pertains.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, You will understand. Therefore, it should be understood that the above-described embodiments are illustrative in all aspects and not restrictive.

Claims (5)

A distance score calculation module for calculating a vector value of a plurality of words existing in the Internet network and calculating a distance score using a distance between a vector of the search word input from the user terminal and a vector of the word for each word;
A real-time issue score calculation module for calculating at least one feature value related to a user behavior for each of the plurality of words using log data, and calculating a real-time issue score using the at least one feature value for each word; And
And a recommendation search word providing unit for providing a recommendation search word including at least one word among the plurality of words to the user terminal according to the distance score and the real-time issue score,
The real-time issue score calculating module calculates,
A feature point calculating unit for calculating a feature point by summing the at least one feature value for each word;
A cluster generator for generating at least one word cluster by grouping a plurality of words having the same retrieval intention among the plurality of words into one word cluster;
Calculating a real-time issue score by calculating a feature score of the at least one word cluster, a sum of feature scores of a plurality of words included in the word cluster, and a feature score of a plurality of non-clustered words, Based on a search behavior pattern. The method according to claim 1,
Wherein the distance score calculating module calculates center coordinates of vectors of the plurality of search words when a plurality of search words input from the user terminal are calculated and calculates a distance between a center of vectors of the plurality of search words and a vector of the word Based on the search behavior pattern. The method according to claim 1,
The distance score calculating module may calculate the distance score based on the search word input from the user terminal,
Wherein a distance between the vectors of the plurality of search words and a vector of the words is calculated, and a distance score is calculated by summing and summing the calculated distances, respectively. The method according to claim 1,
The distance score calculating module may calculate the distance score based on the search word input from the user terminal,
Calculating a distance between the vectors of the plurality of search words and a vector of the words and calculating the distance score using the minimum value or the maximum value among the calculated distances, respectively. The method according to claim 1,
Wherein the recommendation word providing unit is configured to assign a weight to each of the distance score and the real time issue score to calculate a recommendation score and to select a recommendation word including at least one word among the plurality of words according to the recommendation score, Search Query Recommendation System Based on Search Behavior Patterns.

Images