KR20100029581A - Recommended search terms providing system and method for each user and computer readable medium processing the method - Google Patents

Abstract

PURPOSE: A recommended search terms providing system and a method for each user and a computer readable medium processing the method are provided to supply a search term auto-fill list which is suitable for user with higher probability. CONSTITUTION: A cooperated search term filter(31) receives a keyword information for each user and calculates a preference about the key word. The cooperated search term filter calculates the similarity from keyword information and a prediction key word for the each user. A predicted user search term database stores the prediction key word for the each user. A predicted user search terminal selector supplies recommended search term list to a user according from the predicted user search term database.

Description

RECOMMENDED SEARCH TERMS PROVIDING SYSTEM AND METHOD FOR EACH USER AND COMPUTER READABLE MEDIUM PROCESSING THE METHOD}

The present invention relates to a user-specific query recommendation system and method, and in particular, to provide a search word autocompletion list in consideration of a user's personal tendency when a user inputs a search word, so that the user can easily input a user's search word. And a computer readable recording medium capable of implementing the same.

With the rapid development of the Internet, a vast amount of information is provided to users, which makes it increasingly difficult for users to find the information they want. In addition, even if the desired information is found, the time required to find the information (ie, information retrieval time) has increased. As a result, information retrieval using the Internet has become a very common task, and the time required for searching is too large in daily life, and various studies are being conducted to improve the convenience of searching and to improve the efficiency of searching. The result is being applied.

In particular, as the influence of search portals and shopping malls that present search functions as representative functions increases, the intensive efforts to improve user discomfort and improve user satisfaction for the services provided by them by maintaining the market share This is accelerating. As an example, the search term auto-completion service provides a list of search terms in a pre-finished form by inferring the search term that the user wants to search according to the user's input, and reduces the time for entering the search term and provides the search term auto-completion service that informs the exact contents of the search term in advance. It is becoming. For example, when a user enters "Nay" into a search box, the user automatically provides a list of completed search terms such as "Nate" and "Nate On", and the user can enter one of them as a search term.

This query autocompletion service has the effect of reducing the manual input of the user, and a useful search guide if the user does not know the exact information about what they are searching for, or if he or she knows only a few of the search terms. It is used as a search box in most search portals.

However, the current search word autocompletion service calculates the use frequency of most users and provides a list of autocomplete search terms in order of high search frequency. The provision of such a uniform autocomplete search term list may be an easy service when the search term to be searched for appears in the list, but may be an unnecessary service when a list is provided in which the search term does not appear. .

Therefore, a method of introducing personalization into a method of providing a predicted complete search word list based on a user's search word input has been examined.

For example, Korean Patent No. 10-0754768, "A system, a method for providing a personalized recommendation for each user, and a computer-readable recording medium having recorded thereon a program for executing the method," refer to a user's input of a search term. The present invention discloses a technique of extracting and providing a user-specific recommendation word based on the log of the terminal. That is, the technical configuration is intended to provide an autocomplete function based on a user's pre-populated search term, and recommends for the autocompletion by searching a prefix, suffix, etc. according to the user's input. It provides a list of search terms. However, this method is limited to the search terms entered by the user, so personalization may be successful, but when the user enters a query that the user does not know or enters a new search term, the recommendation for the first search term is impossible. Often, the efficiency of providing a search word is lower than that of the existing search word autocompletion service. That is, it merely reduces the time for inputting a search word that is frequently used by the user, and thus it is impossible to achieve the basic purpose of guiding a search word desired by the user among the unspecified search terms.

In order to solve the problems of the conventional uniform search word autocompletion service or the personalized search word autocompletion service, there exist limitations, the object of the present invention is to provide a user's recommendation for a search word that the user has not entered yet. The user's search term recommendation system and method and computer that can implement it can maximize the user's convenience by allowing the user's desired search term to be provided in the search word autocomplete list with higher probability. To provide a recording medium.

Another object of the embodiments of the present invention is to select a prediction query according to the user's disposition by applying a collaborative filtering method using individual search term input information, and to provide an autocomplete search term list to the user based on the selected search terms. By automatically providing search terms according to the user's inclination including the search terms that the user has not entered, the system and method for recommending search terms by user and computer-readable records that can implement the search terms To provide a medium.

Another object of the embodiments of the present invention is to vary the weight of prediction queries for each user based on time with collaborative filtering so as to change interests according to time or to increase the weight of recently searched contents. The present invention provides a user-specific keyword recommendation system and method for enabling personalized search by user, and a computer-readable recording medium capable of implementing the same.

Another object of the embodiments of the present invention is to add a function for filtering abnormal behaviors or meaningless words while inputting a search word to increase the accuracy of the search word autocomplete list, and a computer capable of implementing the search word recommendation system for each user. It is to provide a recording medium that can be read by.

In order to achieve the above object, the user's search term recommendation system according to an embodiment of the present invention receives search term information for each user to calculate a preference for a search term, and calculates a search term similarity from the user's search term use information. A search term collaborative filtering unit to generate a predicted search term; A prediction query database for each user that stores the prediction query for each user generated by the collaboration filtering unit; And a prediction keyword selection unit for each user, which provides a list of recommended search terms for the user from the prediction database for each user according to a user's search word input.

The collaborative filtering unit may further include a search term filtering unit for filtering meaningless words as meaningless search terms or search terms while receiving search word information for each user.

The collaborative filtering unit may further include a search word preference calculator for each user that calculates a preference by applying a weight to the corresponding search word based on a reception time when the search word is received.

The search word preference calculator for each user may determine a user's property by using the user's personal and regional information, and then calculate a preference after assigning a weight to the received search word information for each user property.

The collaborative filtering unit determines two search term combinations that can be selected from two or more different search terms consecutively used by the same user within a preset time interval from the user's search term usage information, as similar search term pairs, and determines the category similarity to which the search term pair belongs. And a search word similarity calculator for calculating search term similarity by applying proportional attribute weights.

The search term similarity calculator may further calculate search term similarity by further applying a time weight that is inversely proportional to the use time interval of the search term pair used by the same user.

The search term similarity calculator may calculate the search term similarity by applying a predetermined search term association degree to the search term pair as an association weight.

The user interface may further include a user determination unit configured to determine a user through at least one of an ID, a network identifier, and a cookie of the collaboration filtering unit and the prediction keyword selection unit for each user, and to provide user identification information.

According to another exemplary embodiment of the present invention, a search term recommendation method for each user receives a search term information for each user to calculate a preference for a search term, and calculates a search term similarity from the search term usage information of the user to generate a prediction term for each user. A collaborative filtering step; And a search word autocompletion list providing step of providing a list of recommended search words for the corresponding user from the prediction database for each user according to a user's search word input.

In addition, the recording medium according to another embodiment of the present invention includes a program for performing the user-specific keyword recommendation method described above.

Such recording media includes all types of recording media on which programs and data are stored so that they can be read by a computer system. Examples include Read Only Memory (ROM), Random Access Memory (RAM), Compact Disk (CD), Digital Video Disk (DVD) -ROM, magnetic tape, floppy disk, optical data storage devices, and the like. It also includes implementations in the form of (eg, transmission over the Internet). In addition, these recording media can be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

According to an exemplary embodiment of the present invention, a system and method for recommending a search word for each user and a computer-readable recording medium capable of implementing the same may allow a user to make a recommendation for a search word that the user has not input, while reflecting the personal characteristics of the user. Search terms can be provided to the search word autocomplete list with a higher probability, thereby maximizing user convenience.

According to an embodiment of the present invention, a system and method for recommending a search word for each user and a computer-readable recording medium for implementing the same may select a search query according to a user's inclination by applying a collaborative filtering method using individual search word input information. By providing a list of autocomplete search terms to the user based on the selected search terms, the search results according to the user's inclination including the search terms that the user has not entered are automatically provided and the user can more accurately recommend the search terms for each user. have.

In accordance with an embodiment of the present invention, a system and method for recommending a search word for each user and a computer-readable recording medium that can implement the same may change interests according to time by varying the weight of the prediction search words for each user based on time along with collaborative filtering. It is possible to change the personalization over time, such as to make different or increase the weight of the recently searched content.

According to an embodiment of the present invention, a system and method for recommending a search word for each user and a computer-readable recording medium embodying the same may add a function to filter abnormal behavior or meaningless words during input of a search word, thereby improving accuracy of the search word autocomplete list. It can increase the effect.

According to an embodiment of the present invention, a system and method for recommending a search word for each user and a computer-readable recording medium capable of implementing the same may calculate a user preference using a user's attributes to consider a user's propensity belonging to a minority group. Even a small number of users can benefit from accurate search suggestions.

The present invention as described above will be described in detail with reference to the accompanying drawings and embodiments.

1 illustrates an example of a search word recommendation system for each user according to an embodiment of the present invention. As shown in FIG. 1, a user inputs a search word and receives an autocomplete search word list based on a text input in real time through a search box. The user of the user according to the browser 10, the web server 20 to communicate with the web browser 10 to provide a search service, and the search word information or real-time search text input text information provided through the web server 20 A search term autocomplete server 30 that provides a list of specialized autocomplete search terms for the user, and a search term usage information (log) database 41 of users, which is the basis for recommending search terms for each user. User-specific prediction query database 42 generated for each user according to each user's preference and search query similarity, and previews the correlation between the search terms. The set search word association is also made including the database 40 including the database 43.

The search term autocompletion server 30 uses search term usage information of users to select search terms having high similarity with search terms received from a specific user and to improve search term preference according to a search term using environment of the specific user. Using the method, a prediction query for the specific user is generated. In addition, the predicted search terms are filtered using the initial consonants, letters, and words of the search word input through the web browser 10, and are arranged in order of high preference, and are provided to the user as an autocomplete search word list.

That is, the query autocompletion server 30 uses an improved collaborative filtering method, not a method of creating an autocompletion search query list among uniform new search queries or an autocomplete search query list only among search terms input by a user. It provides a list of autocomplete search terms in such a way that personalization and search term guides are possible.

Here, collaborative filtering refers to a method of using similarity between users or data when determining data to be recommended to users. In other words, collaborative filtering is a 'user to user collaborative filtering' which determines the recommendation data using the similarity between users and 'item to item collaborative filtering' which determines the recommendation data using the similarity between the data. These are separated by collaborative filtering that merges the two. First, 'collaboration filtering between users' is used to classify users who have similar patterns by using information provided by users (for example, user's personal information and preferences), and then use data usage information of other users included in the same group. Determine the recommendation data of a specific user. That is, cross recommendation is performed between users included in the same group. For example, in the recommendation system using the 'collaboration filtering between users' method, if' User 1 'and' User 2 'are included in' Group 1 'and' User 1 'uses' Data 2', the recommendation system is' Include 'data 2' in the recommendation data of the user 2.

Meanwhile, 'collaboration filtering between items' determines data recommended by a specific user using data pairs used simultaneously by the same user. For example, in a recommendation system using a 'collaboration filtering between items' method, when 'user 1' uses 'data 1' and 'data 2' at the same time, if 'user 2' uses 'data 1' Include 'data 2' in the recommendation data of 'user 2'.

However, this common collaborative filtering method has no consideration of the minority user group, so it is difficult to make accurate recommendations to the users of the minority user group, and even includes inaccurate information, which may lead to inaccurate results. Insensitive to changes in preference tends to lead to inaccurate results.

Therefore, in the present embodiment, a change in propensity over time is considered for a user belonging to a minority user group and a user belonging to a multi-user group, and a personalized search word autocompletion service is provided by a collaborative filtering method in which error information is removed in advance.

The search term autocomplete server 30 generates the prediction query for each user by using the similar search term based on the similarity between the search terms and the preference of the search word considering the user's propensity by applying the improved collaborative filtering method as described above. Even though the search word is not entered by the user, search terms having similarity with the user's previously entered preferred search word are provided as the autocomplete search word.

For example, if a user normally enjoys cooking with vegetables, typing "o" into the search box will provide a list of autocomplete search terms, such as "orange, oyat, oriental dressing ..." If you mainly use search terms for internet shopping or news, you can provide a list of autocomplete searches such as "Auction, OhmyNews, Today's weather ..." by typing "O" in the search box. In other words, different autocompletion search terms are recommended for each user, which is constantly changed according to a user's input.

2 is an embodiment showing a more detailed configuration of the search word auto-completion server 30, and as shown in FIG. 2 to obtain search word information for each user provided from the web server 20 to filter the search words that cause inaccurate results. A user that calculates a user's search query preference using a search query filtering unit 31, a filtered search word, time information at which the search word is received, and environment information (gender, age, interest, local information, etc.) for the user. The search term preference calculator 32 and the search term database 41 storing search term usage information (logs) of the users of the database are selected to search for high similarity terms for the received search term based on the search term use behavior for each user. Matched with the preference information of the corresponding search word by the search word similarity calculator 33 and the user-specific search word preference calculator 32. The search term similarity information by the search term similarity calculator 33 selects search terms to be recommended for the corresponding user, and then generates a predicted search term in the form of data including the preference or similarity related information to the user's prediction search term database. According to the user-specific prediction query generation unit 34 stored in (42) for each user and the real-time input characters (initial, words, words, etc.) that the user enters into the search box through the web server 20 Predictive search query for each user is obtained from the prediction query database for each user 42 is processed in the form of auto-complete search query list information provided by the user to the web browser 10 via the web server 20 prediction query selection unit 35 ).

Since the search word auto-complete server 30 basically operates based on a user's personality, it is necessary to determine a user using a search service. In this case, information about a user (such as user's ID information) is determined according to a user's input. Input identification information, terminal storage information such as cookie information, network identification information such as IP information or MAC (MAC) address information of a LAN card), and the like, and the query filtering unit 31 and the user-specific prediction query selecting unit 35 This is a case where the user determination unit 36 provided in the search word autocomplete server 30 is configured. However, it should be noted that the user determination unit 36 is configured in the web server 20 so that the search word autocomplete server 30 may obtain user determination information through the web server 20.

The configuration of the search word autocomplete server 30 will be described in more detail.

First, the search word filtering unit 31 may increase the accuracy of the collaborative filtering by filtering the meaningless search terms or the meaningless words as search terms while receiving search word information for each user. For example, users who search in English or Korean in English, search for words with a typo, incorrect search terms with no search results, or search words that are too frequently used by users Filter unnecessary search terms to remove unnecessary noise. Of course, in addition to such a case, too many of the same search terms are input within a limited time, or a search word flowing out of a human input condition may be determined as an error and filtered.

The query filtering unit 31 and the user-specific prediction query generator 34 may be selectively applied to the user-specific keyword preference calculator 32 and the keyword similarity calculator 33 to configure a collaborative filtering unit. The user-specific keyword preference calculator 32 and the keyword similarity calculator 33 that perform substantial collaborative filtering may be configured as follows.

First, the search term preference calculator 32 for each user basically applies a weight to a time based on a reception time of a search term received through the search term filter 31, and prefers a search term frequently searched for recently. Calculate your preference so that it is high. If the user accesses through ID input, the user's propensity information of the user who has been surveyed or accumulated can be obtained, and the preference applied to the user's propensity weight can be calculated. If the user is not connected through the ID input, the preference may be calculated by applying the known information (access area information, time, etc.) as the propensity weight. The weight according to the user's inclination means not the simple numerical value but the separation of the result of preference calculation according to a specific property. By checking the propensity of the similar property combination according to the user's property combination such as gender, age, region, and propensity, Make it possible to reflect the propensity of users in the minority group.

The time weight may be applied in various ways. Basically, there may be a method of increasing a weight on a recently searched search term, increasing a preference for a recent search term while decreasing a corresponding time weight as time passes. There may be a method of increasing the preference for search terms that are mainly used in similar time zones by applying a weight according to the difference between the receiving time range of the received search term and the recently received search term, and the frequency of using the same or similar search terms. However, there may be a method of increasing the preference for frequently searched search terms by applying weights according to the use time interval.

The search term similarity calculator 33 selects two search term combinations that can be selected from two or more different search terms consecutively used by the same user within a predetermined time interval from the search term usage information of the users stored in the search term database 41. The search term similarity is calculated by determining a similar search term pair and applying an attribute weight proportional to the category similarity to which the search term pair belongs.

The search term similarity calculator may calculate the search term similarity by further applying a time weight that is inversely proportional to the use time interval of the search word pair used by the same user. In this case, when two or more different search terms are consecutively used by the same user within a preset time interval, it is preferable to determine a combination of two search terms among the search terms as a similar search term pair.

For example, when calculating the similarity with respect to the 'search term 1' entered by the user, the higher the number of users who used the 'search term 1' and the 'search term 2' together at intervals closer to each other, the higher the similarity is calculated. can do.

In addition, although similarity between similar search terms may be obtained through a search term association database 43 that obtains similarity between search terms in advance, it is preferable to apply the method in combination with the collaborative filtering method. That is, the search term similarity may be calculated by applying the search term relevance obtained through the search term relevance database 43 to the search term pair as an association weight.

 Here, the search term similarity calculating unit 33 determines the similarity pair by clustering, and the clustering method is all algorithms used in general collaborative filtering such as Pearson correlation, popularity difference method, and cosine similarity method as is well known. It is possible to use

On the other hand, when obtaining the similarity pairs to calculate the similarity, too many similar content pairs can be determined, so that if the number of users using a specific similar content pair is smaller than the predetermined number of users can be filtered. If the number of users using the corresponding content pair is within a predetermined range of minority users and the attributes of the minority user are equal to or greater than a certain number, the attribute of the minority users and the information of the corresponding content pair are searched by the user's query preference calculator 32. It can also be passed to to be managed as preference information for a specific user group.

Information about similar search terms obtained as a user preference and a similar search word pair for the search word obtained through the user search word preference calculator 32 and the search word similarity calculator 33 may include preference information, similarity information, and time / attribute / Information about propensity weights and the like is generated as a prediction keyword for each user having a two-dimensional or three-dimensional data format and stored in the prediction keyword database for each user 42.

The information stored in the user-specific prediction query database 42 may be updated according to the current state because the weight of the user varies by time and the use of the user's search word is accumulated, and periodically or when there is a user's connection or search. The prediction query for each user may be updated.

The prediction query selector for each user 35 provides a list of recommended search terms for the user from the prediction database for each user according to a user's search term input. It searches for high similarity search terms and generates and delivers a list of autocomplete search terms filtered by letters (first, last, words, end, etc.) in real-time search box. Make it appear.

An example of a method of recommending a user-specific keyword using the above-described configuration will be described with reference to the flowcharts of FIGS. 3 and 4.

FIG. 3 is a flowchart illustrating an example of a process for generating a prediction query for each user. As shown in FIG. 3, when a user inputs a search term through a search box and transmits a search query, the web server or search term auto-completed received the information. The server determines the user. The search word autocompletion server receiving the determined user information and the search word filters the abnormality of the search word through the search word filtering part, and then assigns a weight value according to time or propensity through the collaborative filtering part and calculates a search term preference. The search term similarity is calculated based on the search term usage log. In this case, the time or the attribute weight may also be applied when calculating the search term similarity.

The user's preference calculation value for the search word and the search word similarity calculation may be used to generate the user's prediction search word using the similarity calculation value for the search word pair including the search word, and store it in the user's prediction search word database.

In this way, personalized predicted search term information is obtained for the user, which includes search terms highly relevant to the information that the user has not searched for.

FIG. 4 is a flowchart illustrating an example of providing a user-completed list of autocomplete search terms by using the predicted search terms for each user. As shown in FIG. 4, when a user inputs a search term into a search box through a web browser, The completion server will determine the user. The web server obtains a search word input to the search box in real time and provides the search word to the search word autocomplete server.

The search word autocompletion server selects a prediction query to be recommended from a prediction database for each user, and uses a method of filtering based on spellings (initiality, words, words, etc.) of the search term provided in real time, and user preferences and search terms. Predicted search terms filtered in order of high similarity information are generated and provided in the form of predetermined search word autocomplete list information.

FIG. 5 illustrates an example of a search word autocompletion service using the above-described method. A case in which a specific user inputs 'foreign currency, auction, and apple' as a search word will be described as an example. Search terms such as 'Forex, Auction, Apple' may be regarded as words of interest to the user. In a simple manner, the search word information for the user i becomes Wi = {foreign exchange, auction, apple}.

Here, the result of selecting the high similarity search terms as the search pairs for each search word (that is, the high similarity search word obtained through collaborative filtering, Wr) may be expressed as follows.

Wr (foreign currency) = {bank, money, exchange rate, currency, deposit}

Wr (auction) = {auction, deposit, market, bank}

Wr (apple) = {market, sale, shipping}

Since these search words do not mean only one word, it is also possible to:

Wr (Foreign Exchange, Auction) = {Bank, Deposit}

Therefore, when a user inputs "o" into a search box, an autocomplete search word such as "bank and deposit" may appear in the autocomplete search word list along with "foreign exchange and auction." That is, the recommended search term reflecting the user's interest appears even if the user has never entered it, and the likelihood that the search term is the user's desired search term increases. This may be used to improve input convenience, but may also be used effectively for the purpose of guiding a search word that can satisfy a user's interest in a situation where a large number of search words exist.

Meanwhile, the weighted information according to the time of receiving each of the search words may be reduced by 0.3 since 10 is considered to be lowered after 10 days and one day after the current time. Accordingly, 'foreign' and 'auction' and similar search terms may be preferentially provided, rather than 'apple' and similar search terms. The attenuation form of the time weight according to the time may be applied in various ways such as S-shape, stepped.

In the above described and illustrated with respect to preferred embodiments according to the present invention. The above-mentioned search word recommendation can be applied not only to the Internet environment but also to a text message of a mobile communication terminal, a mobile Internet environment, and the like, and its application range is very wide.

However, the present invention is not limited to the above-described embodiment, and various modifications can be made by those skilled in the art without departing from the gist of the present invention attached to the claims. .

1 is a block diagram showing the configuration of a user-specific keyword recommendation system according to an embodiment of the present invention.

Figure 2 is a block diagram showing the configuration of a search word autocomplete server according to an embodiment of the present invention.

3 and 4 are flowcharts showing the operation of the user-specific keyword recommendation system according to an embodiment of the present invention.

5 is an exemplary search word for explaining an embodiment of the present invention.

** Description of symbols for the main parts of the drawing **

10: web browser 20: web server

30: query autocomplete server 31: query filtering unit

32: query preference calculator by user

33: search term similarity calculating unit 34: user prediction query generation unit

35: prediction query selection unit by user

Claims (24)

A search term collaborative filtering unit configured to receive search term information for each user, calculate a preference for the search term, and calculate a search term similarity from the search term usage information of the users to generate a predicted search term for each user; A prediction query database for each user that stores the prediction query for each user generated by the collaboration filtering unit; And a prediction query selection unit for each user, wherein the prediction query selection unit for each user provides a list of recommended search terms for the corresponding user from the prediction database for each user according to a user's search word input. The system of claim 1, wherein the collaborative filtering unit further includes a search term filtering unit configured to filter meaningless terms as search terms or search terms that are meaningless while receiving search term information for each user. The search word recommendation system for each user according to claim 1, wherein the collaboration filtering unit further includes a search word preference calculator for each user that calculates a preference by applying a weight to the corresponding search word based on a reception time when the search word is received. The system of claim 3, wherein the collaborative filtering unit varies a time reference weight of search terms of the prediction query database for each user over time. The user of claim 3, wherein the user's search word preference calculator determines a user's property using the user's personal and regional information, and then calculates a user's preference after assigning a weighted value to the received user's property. Star Query Recommendation System. The method of claim 1, wherein the collaborative filtering unit determines two search term combinations that are selectable from two or more different search terms consecutively used by the same user within a preset time interval from the user's search term use information as similar search term pairs, and performs the search term. And a search word similarity calculator configured to calculate search word similarity by applying attribute weights proportional to the category similarity to which the pair belongs. The search word recommendation system for each user according to claim 6, wherein the search word similarity calculator calculates the search word similarity by further applying a time weight inversely proportional to the use time interval of the search word pair used by the same user. The system of claim 6, wherein the search term similarity calculator calculates the search term similarity by applying a predetermined search term association degree to the search term pair as an association weight. The system of claim 1, wherein the collaborative filtering unit updates predicted search term data for each user periodically or according to a user's access. The system of claim 1, wherein the prediction keyword selection unit for each user selects a search word having a high preference among search terms in the prediction search database and selects search terms having a high similarity with respect to the search word. The apparatus of claim 1, further comprising a user determination unit configured to determine a user through at least one of an ID, a network identifier, and a cookie of the collaboration filtering unit and the prediction keyword selection unit for each user, and to provide user identification information. User-specific query recommendation system. A search term collaborative filtering step of receiving a search term information for each user to calculate a preference for the search term, and calculating a search term similarity from the search term usage information of the users to generate a predicted search term for each user; And a search word autocomplete list providing step of providing a list of recommended search terms for the user from the prediction database for each user according to a user's search word input. The method of claim 12, wherein the collaborative filtering step further comprises filtering meaningless words as meaningless search terms or search terms while receiving search word information for each user. The method of claim 12, wherein the collaborative filtering step further includes a preference calculation step of applying a weight to the corresponding search word based on a reception time when the search word is received, and calculating a preference for the search word based on the time weight of the search word. How to suggest search terms by user. 15. The method of claim 14, wherein the collaborative filtering step further comprises varying a time reference weight for the search terms of the user-specific predicted search term over time. The method of claim 14, wherein the calculating of the preference further includes determining a property of the user by using the user's personal and regional information, and then calculating a preference after assigning a weight to the received search word information for each user property. How to suggest search terms by user. The method of claim 12, wherein the collaborative filtering step determines two search term combinations that are selectable among two or more different search terms consecutively used by the same user within a preset time interval from the user's search term usage information, and determines the similar search term pairs. And a similarity calculation step of calculating the similarity of the search terms by applying attribute weights proportional to the similarity of the category to which the search term pair belongs. The search word recommendation for each user according to claim 17, wherein the calculating of the similarity further comprises calculating a search term similarity by further applying a time weight inversely proportional to the use time interval of the search term pair used by the same user. Way. The method of claim 17, wherein the calculating of the similarity comprises calculating search term similarity by applying a correlation weight of the pair of search terms obtained through a separate search term relevance database as an association weight. The method of claim 12, wherein the collaborative filtering step further comprises updating predicted user-specific search word data generated periodically or according to a user's access. The method of claim 12, wherein the providing of the search word autocomplete list comprises generating a recommendation search list by selecting search words having high preference and search terms having high similarity to the search word among the search terms in the prediction search database. How to suggest search terms by user. The method of claim 12, wherein the providing of the search word autocompletion list further comprises filtering the recommendation search word list according to a real-time input character according to a user's search word input and providing the result as an autocomplete search word list. How to suggest search terms by user. The user-specific search word of claim 12, further comprising a user determination step of determining a user through at least one of a user ID, a network identifier, and a cookie before the collaborative filtering step and the search word autocomplete list providing step. Recommended way. A computer-readable recording medium containing a program capable of performing the method of claim 12.

Images