KR20200125532A - Item recommendation method of using degree of association and sequence between unit of language and using item - Google Patents

Abstract

The present invention relates to an item recommendation method using a degree of association and a sequence between a unit of a language and a using item for recommending items with high purchase potential using a vector model-based inference method based on a degree of association between words and words, not simply a matching method expressed in units of a language by drawing up a novel perception system through research on emotions and tastes of people, time, place, and occasion (T.P.O), and the characteristics of various categories (color, texture, etc.) based on a language used by a user in daily life based on units of a language, user preference, experience, and/or a purchase history using a machine learning system and applying and implementing machine learning technology and natural language understanding technology.

Description

Item recommendation method using association and order between language units and items used {ITEM RECOMMENDATION METHOD OF USING DEGREE OF ASSOCIATION AND SEQUENCE BETWEEN UNIT OF LANGUAGE AND USING ITEM}

The present invention relates to an item recommendation method using a degree of association and an order between a language unit and a used item, and more specifically, based on a language unit and a user's preference or experience and/or purchase history and order using a machine learning system. Therefore, based on the languages used by the user in daily life, a new recognition system is conceived through research on human emotions and tastes, TPO (Time, Place, Occasion), and characteristics of various categories (color, texture, etc.), and machine learning. By applying technology and natural language understanding technology, it is intended to recommend items with high purchase potential using a vector model-based inference method based on the degree of association between words and words, not simply a matching method expressed by keywords. It relates to an item recommendation method using a degree of association and order between language units and items to be used.

In general, modern information users (web users) have to select the information they want in an environment in which a large amount of information and various types of information are simultaneously provided. In addition, since the information required by the user has a variety of formats according to the source of the information, a more intelligent and automated system is required for comprehensive analysis and utilization.

In order to satisfy these requirements, a system that analyzes and recommends web users' search patterns was used.

A typical application for research on the search pattern of web users is to analyze session data of users who access their server from the server side and use it for advertisement or user-customized web page production.

This analysis result is recognized as one of the most important management strategies for companies that have Internet users as their main customers along with Collaborative Filtering technology. In other words, users' search results (eg, search page, search time, search order, transaction trend, etc.) recorded in the server are used as one of the main sources of value-added. The analyzed data is used to infer a web page to be visited later by the user or to infer a product of interest, and as a result, the application is being made in the direction of recommending a product with high purchase potential suitable for the user's characteristics. In addition, there are many attempts to analyze the user's search pattern on the client side in the form of a proxy server or plug-in, which has the advantage of utilizing more various information than the analysis on the server side.

However, there is a problem in that the existing user preference analysis system and recommendation service cannot use various information and cannot overcome the limitations of using only the information given in their respective domains.

1. Korean Utility Model Publication No. 20-0426027, "User-defined information retrieval system" (announced on September 18, 2006)

An object of the present invention was devised to solve the above-described problem, and a language unit for recommending similar items with high purchase possibility based on a language unit and a user's preference or experience and/or purchase history and order, and It is intended to provide a method of recommending items using a degree of association and order between items used.

An item recommendation method using a degree of association and an order between a language unit and a use item according to the present invention is an item recommendation method using a degree of association and an order between a language unit and a use item performed by an item recommendation device, and Obtaining a degree of association between language units-items set in advance for each category; Obtaining a usage history of the category by a control unit of the item recommendation device; Inputting, by a control unit of the item recommendation device, language unit-relevance lists corresponding to the usage details into a machine learning system, and obtaining a recommended language unit-relevance list; And providing, by a control unit of the item recommendation device, recommended items to the terminal in an order corresponding to the recommendation-related language unit-relevance list.

The item recommendation method using a degree of association and order between a language unit and an item to be used according to the present invention has an advantage of being able to recommend information that is efficient, effective, and highly likely to be purchased by using the degree of association and order between the language unit and the item used. .

1 is a schematic block diagram of an item recommendation system using an association diagram and an order between a language unit and a used item according to an embodiment of the present invention.
FIG. 2 is a block diagram showing the configuration of the terminal shown in FIG. 1;
A configuration block diagram of the item recommendation device shown in FIG. 1 of FIG. 3.
4A to 4C are flowcharts of an item recommendation process through an item recommendation interface according to an embodiment of the present invention.
5 is a flowchart of an item recommendation process according to an embodiment of the present invention.
6 is a detailed flowchart of some of the steps of FIG. 5;
7 is a detailed flowchart of some of the steps of FIG. 6.
Figure 8 is a detailed flow chart of some other steps of Figure 6;
9 is a flowchart of an item recommendation process according to another embodiment of the present invention.
10 is an exemplary diagram of a stored object-language unit association diagram according to an embodiment of the present invention.
11 is an exemplary diagram of a basic language unit-language unit association diagram according to an embodiment of the present invention.
12 is a detailed flowchart of some other steps of FIG. 5.
13 is a flowchart of an item recommendation process according to another embodiment of the present invention.
14 is a flowchart of an item recommendation process according to another embodiment of the present invention.
15 is a detailed flowchart of some of the steps of FIG. 13 according to an embodiment of the present invention.
16 is a detailed flowchart of some of the steps of FIG. 13 according to another embodiment of the present invention.
Figure 17 is a detailed flow chart of some of the steps of Figure 13 according to another embodiment of the present invention.
18 is a diagram showing a term hierarchy according to an embodiment of the present invention.
19 is a flowchart of a process of installing a virtual input interface program according to an embodiment of the present invention.
20 is a flowchart illustrating a method of setting a degree of association between language units, usage details, and order according to an embodiment of the present invention.
21 is a diagram illustrating an example of a machine learning system used to set a degree of association and order between a language unit and a use item according to an embodiment of the present invention.
22 is an exemplary view showing main language units for each language unit list based on a language unit associated with a selected item according to an embodiment of the present invention.
23 is an exemplary view showing main language units for each language unit list based on a language unit associated with a selected item according to another embodiment of the present invention.
24 and 25 are exemplary views showing an average correlation criterion of major language units for each list according to an embodiment of the present invention.
26 is an exemplary view showing an average correlation criterion of major language units for each list according to another embodiment of the present invention.
27 and 28 are exemplary views showing language units and language unit categories of major language units for each list according to an embodiment of the present invention.
29 is an exemplary view showing language units and language unit categories of main language units for each list according to another embodiment of the present invention.
30 is an exemplary view showing a language unit adjustment interface corresponding to a clicked list according to an embodiment of the present invention.
31 is an exemplary view showing an interface for adjusting all language units corresponding to a clicked list according to an embodiment of the present invention.
32 is an exemplary view showing an interface for adjusting a language unit corresponding to a clicked list according to another embodiment of the present invention.
33 is an exemplary view showing an interface for adjusting all language units corresponding to a clicked list according to another embodiment of the present invention.
34 is a diagram illustrating an example of a recommendation item recommended by an item recommendation method using a degree of association and an order between a language unit and a use item according to the present invention.

Hereinafter, an item recommendation method using a degree of association and order between a language unit and a use item according to the present invention will be described in more detail through detailed description of embodiments with reference to the drawings. In describing the present invention, if it is determined that a detailed description of a related known technology or configuration may unnecessarily obscure the subject matter of the present invention, a detailed description thereof will be omitted. In addition, terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a client, an operator, or a user. Therefore, the definition should be made based on the contents throughout this specification.

The same reference numerals refer to the same elements throughout the drawings.

Hereinafter, in the present application, the language unit (which may also be referred to as "keytalk") refers to a basic information search unit used in an Internet service based on an interactive interface such as voice, and is expressed in the form of a corpus. It is not a simple word matching method in the document, but based on the proximity and similarity between various expressions, and attribute data such as color and texture, and is generally composed of one heading and multiple sub-attribute words and weights that are highly related to this heading. Since it is expressed as a configured function, it is assumed that it is designed in a structure capable of inferring the taste and intention of the user from the speech of the search request and recommending the corresponding result.

1 is a schematic block diagram of an item recommendation system using an association diagram and an order between a language unit and a used item according to an embodiment of the present invention.

Referring to FIG. 1, an item recommendation system using a degree of association and an order between a language unit and a use item according to an embodiment of the present invention may include a communication network 150, a terminal 200, and an item recommendation device 300. have.

Here, the terminal 200 is, for example, a smart phone, PDA, tablet PC, notebook computer, laptop computer, personal computer, and other electronic devices or similar devices capable of performing communication, receiving user input, and outputting a screen. Can be implemented.

In addition, the item recommendation device 300 may be implemented as, for example, a workstation, a server, a general computer, an electronic device capable of performing other communication, or a similar device.

In addition, the terminal 200 and the item recommendation device 300 are connected through the communication network 150 and communicate with each other through the communication network 150.

In addition, the communication network 150 is, for example, LTE (Long Term Evolution), LTE-A (LTE-Advanced), WI-FI, LAN (Local Area Network), WAN (Wide Area Network), CDMA (Code Division Multiple Access). ), Time Division Multiple Access (TDMA), Wireless Broadband (WiBro), Global System for Mobile Communications (GSM), and others. It may be implemented using at least some of the communication methods developed in the past, present, or available in the future. Hereinafter, for convenience, it will be described as if the terminal 200 and the item recommendation device 300 directly communicate with each other without referring to the communication network 150.

Detailed operations and configurations of the terminal 200 and the item recommendation device 300 will be described later with reference to FIGS. 2 to 8.

2 is a block diagram showing the configuration of the terminal shown in FIG. 1.

Referring to FIG. 2, the terminal 200 may include an input unit 210, a display unit 220, a communication unit 230, a storage unit 240, and a control unit 250.

Here, the input unit 210 converts a user's input operation into an input signal and transmits it to the control unit 250. The input unit 210 may be implemented as, for example, a keyboard, a mouse, a touch sensor on a touch screen, a touch pad, a keypad, a voice input, and other input processing devices that are possible in the present, in the past, or in the future. The input unit 210 may receive, for example, a user's item recommendation request input and transmit it to the control unit 250.

In addition, the display unit 220 outputs a screen according to the control of the controller 250. The display unit 220 may be implemented as, for example, an LCD (liquid crystal display), an LED (light emitting diode), an OLED (organic light emitting diode), a projector, and other display devices that are possible in the present, in the past, or in the future. The display unit 220 may display, for example, an interface page for item recommendation or an item recommendation result page.

In addition, depending on the embodiment, instead of the screen output, a component using a different method for transmitting information to the user such as voice output or vibration may be used instead of the display unit 220.

In addition, the communication unit 230 exchanges data with the item recommendation device 300 and/or other external devices. The communication unit 230 transmits the data received from the item recommendation device 300 to the control unit 250. In addition, the communication unit 230 transmits data to the item recommendation device 300 under the control of the controller 250. The communication technology used by the communication unit 230 may vary depending on the type of communication network 150 or other circumstances.

Also, the storage unit 240 stores data under the control of the controller 250 and transmits the requested data to the controller 250.

In addition, the controller 250 controls the overall operation of the terminal 200 and each component. The control unit 250 transmits an item recommendation request and other data to the item recommendation device 300 according to the information input from the input unit 210 as described later, and according to the page information received from the item recommendation device 300 The result page and/or the interface page are displayed through the display unit 220. The operation performed by the control unit 250 may be distributedly processed by several physically separated computing devices. Some of the operations performed by the controller 250 are performed by the first server and other operations are performed by the second server. In this case, the control unit 250 may be implemented as a total number of physically separated computing devices.

Here, the storage unit 240 may be implemented as a total of physically separated storage devices.

In addition, when the controller 250 or the storage unit 240 is implemented as a total of several devices that are physically separated, communication between several devices may be required. Here, for the sake of simplicity, description will be made on the assumption that the storage unit 240 or the control unit 250 is implemented as one object.

In addition, when the terminal 200 transmits and receives data, it may be expressed that the communication unit 230 transmits and receives data according to the control of the controller 250 according to the viewpoint, and the controller 250 controls the communication unit 230 to transmit data. It can also be expressed as sending and receiving.

Specific operations of each component of the terminal 200 will be described later with reference to FIGS. 4A to 8.

3 is a block diagram showing the configuration of the item recommendation device shown in FIG. 1.

Referring to FIG. 3, the item recommendation device 300 may include a communication unit 310, a control unit 320, and a storage unit 330.

Here, the communication unit 310 exchanges data with the terminal 200. The communication unit 310 transmits the data received from the terminal 200 to the control unit 320. In addition, the communication unit 310 transmits data to the terminal 200 under the control of the controller 320. The communication technology used by the communication unit 310 may vary depending on the type of communication network 150 or other circumstances.

In addition, the storage unit 330 stores data under the control of the controller 320 and transmits the data requested from the controller 320 to the controller 320.

In addition, the controller 320 controls the overall operation of the item recommendation device 300 and each component. When receiving an interface page request, an item recommendation result page request, and other data through the communication unit 310, the control unit 320 loads necessary data from the storage unit 330 and generates page information. The page information is transmitted to the terminal 200 through the communication unit 310.

Here, when the item recommendation device 300 transmits and receives data, it may be expressed that the communication unit 310 transmits and receives data according to the control of the controller 320 according to the viewpoint, or the controller 320 controls the communication unit 310 It can also be expressed as transmitting and receiving data. In addition, according to an embodiment, data for providing information may be transmitted and received by voice or other methods instead of a page for visually providing information.

Detailed operations of each component of the item recommendation device 300 will be described later with reference to FIGS. 4A to 8.

4A to 4C are flowcharts illustrating an item recommendation process through an item recommendation interface according to an embodiment of the present invention.

4A to 4C, an item recommendation process through the item recommendation interface is as follows.

First, in step S401, a degree of association between language units and items is set by the control unit 320 of the item recommendation device 300. This process may be performed by a method described later with reference to FIGS. 5 to 9, may be set by an administrator, or may be set in another method.

Thereafter, in step S402, an experience item list including the user's experience item is obtained by the control unit 320 of the item recommendation device 300. Here, the experience item is, for example, a purchase of the product, a preference mark (e.g., like), a visit record (e.g., a photo held by the terminal, a photo uploaded by the user, or location information of the terminal) Analysis, etc.), items with selection/inquiry records, and can be limited according to item categories (travel destinations/movies/broadcast programs). In addition, the experience item may be limited to a period within a certain period from the present to the past according to an embodiment.

Steps S403 to S407 below may be omitted when servicing only one item category.

Thereafter, in S403, the control unit 320 of the item recommendation device 300 generates item category selection interface page information including a language unit list together with an experience item. Here, the item category selection interface page information is information necessary to generate an item category selection interface page. In addition, the item category selection interface page is a page for inducing a user's input, receiving the user's input, and transmitting it to the item recommendation device 300. For example, the item category selection interface page information may be in the form of an HTML document or other markup language document. In addition, according to an embodiment, the terminal 200 has the form information of the item category selection interface page in advance, and only items corresponding to the content may be transmitted from the item recommendation device 300 to the terminal 200. Hereinafter, for convenience, it is assumed that interface page information or other page information is delivered in an HTML document format. However, it is not intended to limit the present invention to this.

Thereafter, in step S404, the item category selection interface page information is transmitted to the terminal 200 by the communication unit 310 of the item recommendation device 300.

Thereafter, in step S405, the control unit 250 of the terminal 200 constructs an item category selection interface page using the item category selection interface page information, and the display unit 220 of the terminal provides the user 400 with the item category selection interface page. Is displayed. Here, the item category selection interface page may include, for example, an interface through which the user 400 requests item recommendation, inputs and/or selects a language unit for item recommendation, and sets for other item recommendations. have.

Thereafter, in step S406, the input unit 210 of the terminal 200 receives the item category selection input of the user 400 inputted through the item category selection interface page and transmits it to the controller 250. Here, the virtual user's experience is selected.

Thereafter, in step S407, the communication unit 230 of the terminal 200 transmits item category selection input information for identifying the user's item category selection input to the item recommendation device 300 under the control of the controller 250. .

Thereafter, in step S408, the control unit 320 of the item recommendation device 300 obtains a language unit/relevance pair corresponding to the user's experience item list. Here, the language unit/relevance pair is associated with each experience item in the experience item list including the user's experience item acquired in step S402, and is obtained based on a threshold value. In addition, the threshold value may be, for example, a maximum number or a minimum degree of association.

For example, according to an embodiment in which the threshold value is the maximum number, the control unit 320 selects the maximum number of language units (for example, 10) of language units in which a degree of association is set for a specific experience item in the experience item list The degree of relevance (the degree of relevance to the relevant experience item) can be selected in the order of high, and acquired/stored as a language unit-relevance pair.

This process can be performed for all experience items included in the experience item list. If i1 to i10 are included in the experience item list, and the language unit-related degree pair corresponding to i1 is p1_1 to p1_10, p2_1 to p1_10 .... p10_1 to p10_10 may be acquired/stored through the same process. The language unit-relationship pair obtained in this way may be used to obtain a language unit-user association in step S431.

For example, according to an embodiment in which the threshold is the maximum number, the control unit 320 has the minimum degree of association (the degree of relevance to the corresponding experience item) among language units in which a degree of relevance is set for a specific experience item in the experience item list. For language units with a degree of association or higher, a language unit-association degree pair can be obtained/saved. This process can be performed for all experience items included in the experience item list.

The processes of steps S410 to S430 may be provided when additional management is required by the administrator or may be omitted. In addition, in the processes of S410 to S430, an interface may be provided to the service manager, not the user.

Thereafter, in step S410, the control unit 320 of the item recommendation device 300 generates a "main language unit for each language unit list" screen, a web chart, and a scattered chart for each language unit list.

Thereafter, in step S411, the communication unit 310 of the item recommendation device 300 transmits a screen, a web chart, and a distributed chart to the terminal 200 under the control of the controller 320. Deliver.

Thereafter, in step S412, the display unit 220 of the terminal 200 displays a "main language unit for each language unit list" screen, a web chart, and a distributed chart to the user 400.

Thereafter, in step S413, the input unit 210 of the terminal 200 receives the enlarged input of the user 400 in the scatter chart.

Thereafter, in step S414, the communication unit 230 of the terminal 200 transmits the enlarged similar language unit information according to the enlarged input of the user 400 to the item recommendation device 300 under the control of the control unit 250.

After that, in step S415, the interface page information in units of similar languages is generated by the control unit 320 of the item recommendation device 300.

Thereafter, in step S416, the interface page information in units of similar languages is transmitted to the terminal 200 by the communication unit 310 of the item recommendation device 300.

Thereafter, in step S417, the control unit 250 of the terminal 200 constructs a similar language unit interface page using the similar language unit interface page information, and the display unit 220 of the terminal provides the user 400 with the similar language unit interface page. Is displayed.

Thereafter, in step S418, the input unit 210 of the terminal 200 receives the edit input for the list of specific language units of the user 400 and transmits the input to the control unit 250.

Thereafter, in step S419, the communication unit 230 of the terminal 200 enters an edit input for a specific language unit list capable of identifying an edit input for a specific language unit list of the user 400 under the control of the control unit 250 The information is transmitted to the item recommendation device 300.

Thereafter, in step S420, interface page information for adding/deleting a language unit corresponding to a specific language unit list is generated by the control unit 320 of the item recommendation device 300.

Thereafter, in step S421, interface page information for adding/deleting language units corresponding to the specific language unit list is transmitted to the terminal 200 by the communication unit 310 of the item recommendation device 300 under the control of the controller 320. Deliver.

Thereafter, in step S422, the control unit 250 of the terminal 200 configures an interface page for adding/deleting language units by using the interface page information for adding/deleting language units, and the display unit 220 sends to the user 400. Displays the interface page for adding/deleting language units corresponding to the specific language unit list.

Thereafter, in step S423, the input unit 210 of the terminal 200 adds a language unit corresponding to the specific language unit list requested to be added/deleted through the interface page for adding/deleting the language unit corresponding to the specific language unit list. /Delete.

Thereafter, in step S424, the input unit 210 of the terminal 200 receives an input of “selecting a language unit” from the user 400 and transmits it to the control unit 250.

Thereafter, in step S425, the communication unit 230 of the terminal 200 transmits input information of “selecting a language unit” to identify an input of “selecting a language unit” according to the control of the controller 250. ).

Thereafter, in step S426, the control unit 320 of the item recommendation device 300 generates the addition/deletion interface page information for the entire language unit.

Thereafter, in step S427, the communication unit 310 of the item recommendation device 300 transmits the addition/deletion interface page information for the entire language unit to the terminal 200 under the control of the controller 320.

Thereafter, in step S428, the control unit 250 of the terminal 200 configures the add/delete interface page using the add/delete interface page information, and the display unit 220 adds/deletes the entire language unit to the user 400. Display the interface page. Here, language units are grouped and displayed according to a list.

Thereafter, in step S429, the input unit 210 of the terminal 200 receives the add/delete input for the entire language unit of the user 400 and transmits the input to the control unit 250.

Thereafter, in step S430, the communication unit 230 of the terminal 200 recommends the addition/deletion input information for the entire language unit to identify the addition/deletion input for the entire language unit under the control of the controller 250. To the device 300.

Thereafter, in step S431, a language unit between the user and each language unit based on a language unit-relevance pair between all language units corresponding to the experience item list by the control unit 320 of the item recommendation device 300 -Acquire user relevance. According to some embodiments, the controller 320 includes a cumulative value of a degree of association corresponding to a specific language unit with respect to the entire language unit-relevance degree pair corresponding to the experience item list and a language unit between the user and the specific language unit. Language unit-user association can be set so that the user association has a positive correlation. For example, item 1, item 2, item 3, ..., item k are included in the experience item list, and the set of language unit-relevance pairs corresponding to item n is {(T1, W1_n), (T2 , W2_n), ..., (Tj, Wj_n)}, the cumulative value of the degree of association corresponding to the specific language unit Tm may be ∑ _n=1..k (Wm_n). Where Wm_n is the degree of association between item n and a specific language unit Tm. According to some embodiments, when the association of item n and a specific language unit Tm is 0 or less than or equal to a specific reference value, the (Tm, Wm_n) pair may be omitted without being stored. The controller 320 may not store information corresponding to the language unit-relationship pair (Tm, Wm_n) when the association of item n and the specific language unit Tm is 0 or less than a specific reference value. Thereafter, if information corresponding to the language unit-relationship pair (Tm, Wm_n) is not stored, the controller 320 may treat the item n as having a degree of association of 0 corresponding to the specific language unit Tm.

According to some embodiments, the controller 320 determines that the cumulative value of the degree of association corresponding to the specific language unit with respect to the entire language unit-relevance pair corresponding to the experience item list is a language unit between the user and the specific language unit- The language unit-user association may be set so that the degree of user association or the cumulative value of the association degree corresponding to the specific language unit is proportional to the language unit-user association between the user and the specific language unit.

Thereafter, in step S432, the control unit 320 of the item recommendation device 300 generates user recommendation item list information based on a language unit-user association.

For example, the control unit 320 is a value obtained by accumulating individual values having a positive correlation to the language unit-user association degree for a specific language unit and the language unit-item association degree for the specific language unit for all language units Item-user association can be set to have a positive correlation with.

In a specific embodiment, the controller 320 multiplies a language unit-user association degree for a specific language unit and a language unit-item association degree for a specific language unit, and a positive correlation with a value accumulated for all language units. Item-user association can be set to have a relationship.

In a specific embodiment, the controller 320 is a value obtained by multiplying a language unit-user association degree for a specific language unit and a language unit-item association degree for the specific language unit as a value accumulated for all language units. You can set the degree of association.

For example, if the language unit-user association between a specific language unit i and the user is U _i and the language unit-item association between a specific item I _j and the language unit i is R _ij, then the specific item I _j and the user -User relevance can be expressed as ∑ _i=0...k-1 f(R _ij , U _i ). Here, k is the number of language units having a non-zero language unit-item association in relation to a specific item I _j . Here, f(x, y) may be, for example, a value obtained by multiplying x and y, or may be another operation in which x and y and f(x, y) have a positive correlation. In the modified example, the item-user association and the cumulative value of ∑ _i=0...k-1 f(R _ij , U _i ) are not just cumulative (∑ _i=0...k-1 ). Other equations, including proportional operations, logarithmic operations, etc., can be used to find the item-user association.

The controller 320 may generate a user recommended item list according to the item-user association.

Thereafter, in step S433, the communication unit 310 of the item recommendation device 300 transmits the user recommendation item list information to the terminal 200 under the control of the controller 320.

Thereafter, in step S434, the control unit 250 of the terminal 200 constructs a user recommendation item using the user recommendation item list information, and the display unit 220 displays the user recommendation item to the user 400.

In the embodiments of FIGS. 4A to 4C, it is assumed that a page in a visual form is provided to the user, but it is also possible to provide an interface or result information through voice. In this case, an audio output unit may be used instead of the display unit 220. Instead of the visual/audible method, another type of interface method that is currently possible or will be possible in the future may be used in the relationship with the user 400. In this case, the item recommendation device 300 may provide the terminal 200 with information converted in a different manner instead of page information to fit the interface method.

In the embodiment after FIG. 5, the user 400 wants to be provided with information on an object of a specific field of interest that the user 400 is interested in. However, it is not intended to limit the present invention to this.

Here, the field of interest may be, for example, an object type. For example, if the field of interest is'celebrity', the object corresponding to this field of interest may include'Yoo Si O','Yoo Jae O', and'Taylor Sw O'. For example, if the field of interest is'movie', objects corresponding to this field of interest may include'Dunkir O','Spider O Homeker O','Super Boat O 3', and the like. For example, if the field of interest is'broadcasting program', objects corresponding to this field of interest may include'Infinite OO','American AOO','OO of the Throne', and the like.

In the following embodiments, a document is collected to evaluate the relationship (relevance/weight, etc.) between language units. Collected documents may all be evaluated as having equal value, or newer documents may be evaluated as having higher value. That is, the association between the age of the document at the time of evaluation and the language units appearing in the document may have a negative correlation.

Even in the process after FIG. 5 to be described later, the value may be evaluated differently according to the freshness of the document. For example, a case where two language units appear together in a document one day after the evaluation point can be evaluated with a 10 times higher association than a case where two language units appear together in a document 10 days after the evaluation point. The age of a document may be evaluated in units of seconds/minutes/hours, or in units of days/months/years. For the document evaluated before the age of the document is reflected, the control unit 320 divides the partial relevance value by the age of the document, extracts the partial relevance reflecting the document age, and accumulates the partial relevance to the two languages. The degree of association between units can be extracted.

The creation time of a document to check the age of a document can be determined using, for example, the posting time and/or metadata included in the document, or through periodic crawling, when a new document not found in the previous crawl is found. In this case, it may be determined that a new document was added at the time of a new crawl.

5 is a flowchart of an item recommendation process according to an embodiment of the present invention.

In step S510, an object-language unit association diagram corresponding to an object item and a representative attribute language unit pair is stored in the storage unit 330 by the control unit 320. The process of step S510 may be performed according to, for example, some of the embodiments of FIGS. 6 to 8, a process similar thereto, or a process corresponding thereto. Depending on the embodiment, the process of step S510 may be performed as an input of an administrator or may be performed by receiving an object-language unit association degree determined in an external system through a network or a storage medium.

Meanwhile, the stored object-language unit association is as shown in FIG. 10 showing an example of the stored object-language unit association according to an embodiment of the present invention. In the embodiment of FIG. 10, there are all m items (i ₁ to i _m ) and n representative attribute language units are all n (k ₁ to k _n ). For example, the degree of association between object item i ₅ and representative attribute language unit k ₃ is w _5,3 .

In addition, step S510 is as follows with reference to FIG. 6 showing a detailed procedure of step S510 of FIG. 5.

Referring to FIG. 6, in step S610, a representative attribute language unit candidate set is extracted from the first set of documents by the control unit 320. In this case, the controller 320 may collect, for example, language units frequently appearing in a document corresponding to an interest field among the first set of documents as a representative attribute language unit candidate set.

In addition, step S610 is as follows with reference to FIG. 7 showing a detailed procedure of step S610 of FIG. 6.

Referring to FIG. 7, the controller 320 includes language units appearing in a document, such as an object language unit representing an object item belonging to a specific interest field, and language units appearing in a document, such as a field language unit indicating a specific interest field. It may be set as a first attribute language unit candidate set and a second attribute language unit candidate set.

For example, if the field of interest to provide item recommendation service is a celebrity, the field language unit may include'celebrity','celebrity','movie actor','star','celebrity','celeb', etc. . The field language unit may be set by an administrator or recommended/set by the controller 320. After acquiring some field language units, the controller 320 may recommend/set a similar language unit, which is analyzed to have a degree of association with the field language units greater than or equal to a preset value, as an additional field language unit.

In addition, if the field of interest to provide the item recommendation service is a famous person, the object language unit may be an individual person belonging to the field of interest. For example, "Yoo Jae O", "Taylor Swew OO", and "Stefan OO" may be object language units corresponding to celebrities in the field of interest.

Here, to describe the relationship between the field language unit and the object language unit, for example, the field language unit may correspond to a property or type of the object language unit. If the field language unit represents a set, the object language unit may represent the elements belonging to that set.

In addition, the object language unit may be set by the administrator or may be selected in a similar manner to the field language unit. Depending on the embodiment, the controller 320 may analyze the context of the collected documents and select language units determined as elements belonging to the set indicated by the field language unit as the object language unit.

Also, the popular object language unit and the non-popular object language unit may be classified according to the search/collection amount of the corresponding object language unit. The controller 320 may search/collect a document including each object language unit, set an object language unit having a collection amount equal to or greater than a specific threshold value as a popular object language unit, and set the remaining object language units as an unpopular object language unit.

In addition, the language unit of the popular field and the language unit of the non-popular field may be classified according to the amount of search/collection of the language unit of the corresponding field. The controller 320 may search/collect a document including each field language unit, set a field language unit having a collection amount equal to or greater than a specific threshold value as a popular field language unit, and set the remaining field language units as an unpopular field language unit. However, a threshold value for classifying a popular object language unit and a non-popular object language unit and a threshold value for classifying a popular field language unit and an unpopular field language unit may be different values. Hereinafter, for convenience, the popular object language unit and the popular field language unit are collectively referred to as a popular field object language unit. In addition. For convenience, the non-popular object language unit and the non-popular field language unit are collectively referred to as the non-popular field object language unit.

In addition, in a modified embodiment, only the popular field language unit or the popular object language unit may be used instead of the popular field object language unit. In a modified embodiment, only a non-popular field language unit or a non-popular object language unit may be used instead of the non-popular field object language unit.

In step S710, language units appearing in the same document as the popular field object language unit are set by the control unit 320 as a first attribute language unit candidate set.

Here, the controller 320 may search/collect documents including a popular field object language unit and set language units included in the collected document as a first attribute language unit candidate set.

According to another embodiment, the controller 320 may exclude a field language unit and an object language unit from among language units included in the collected document from the first attribute language unit candidate set. In addition, the controller 320 may exclude a pre-set meaningless language unit, for example, a survey/article, from the first attribute language unit candidate set. According to another embodiment, the controller 320 may include, in the first attribute language unit candidate set, a language unit registered in a preset dictionary among language units included in the collected document.

According to another embodiment, the control unit 320 includes a popular field object language unit.

Search/collect documents, and in the collected documents, the popular field object language unit or the language unit

Language units arranged within a preset distance from the included sentence may be included in the first attribute language unit candidate set. According to another embodiment, the control unit 320 searches/collects a document including a language unit of a popular field object, analyzes a context, and selects the language units used for formulating/descripting the language unit of the popular field object. It can also be included in the attribute language unit candidate set.

In addition, the distance between the language units or the distance between the language unit and the sentence is, for example, the number of sentences positioned between two language units or a language unit and a sentence, the number of words positioned between two language units or a language unit and a sentence, It may be determined based on one or more of the number of words positioned between two language units or a language unit and a sentence, and the number of letters between two language units or a language unit and a sentence. The control unit 320 may first perform a morpheme analysis to analyze a language unit.

In step S720, language units appearing in the same document as the non-popular field object language units are set as the second attribute language unit candidate set by the controller 320.

Here, the controller 320 may search/collect a document including an unpopular field object language unit, and set language units included in the collected document as a second attribute language unit candidate set.

According to another embodiment, the controller 320 may exclude a field language unit and an object language unit from among language units included in the collected document from the second attribute language unit candidate set. In addition, the controller 320 may exclude a pre-set meaningless language unit, for example, a survey/article, from the second attribute language unit candidate set. According to another embodiment, the controller 320 may include, in the second attribute language unit candidate set, a language unit registered in a preset dictionary among language units included in the collected document.

According to another embodiment, the control unit 320 searches/collects a document containing an unpopular field object language unit, and places it within a preset distance from the unpopular field object language unit or a sentence containing the language unit in the collected document. The configured language units may be included in the second attribute language unit candidate set. According to another embodiment, the control unit 320 searches/collects a document including a non-popular field object language unit, analyzes the context, and sets the second language units used for formulating/description It can also be included in the attribute language unit candidate set.

In addition, the distance between the language units or the distance between the language unit and the sentence is, for example, the number of sentences positioned between two language units or a language unit and a sentence, the number of words positioned between two language units or a language unit and a sentence, It may be determined based on one or more of the number of words positioned between two language units or a language unit and a sentence, and the number of letters between two language units or a language unit and a sentence. The control unit 320 may first perform a morpheme analysis to analyze a language unit.

In step S730, language units belonging to both the first attribute language unit candidate set and the second attribute language unit candidate set may be set as the representative attribute language unit candidate set by the controller 320. That is, language units that are used as expressions for modifying popular field object language units and also used as expressions for non-popular field object language units may be collected as representative attribute language unit candidate sets.

According to another embodiment, in step S610, the controller 320 may include language units appearing together with the object language unit and/or the field language unit in the representative attribute language unit candidate set regardless of whether or not they are popular or unpopular.

Returning to FIG. 6, in step S620, two or more lower language units associated with each representative attribute language unit included in the representative attribute language unit candidate set are extracted from the second set of documents by the control unit 320.

The second set document used for extracting the lower language unit in step S620 and the first set document used for extracting the representative attribute language unit candidate set in step S610 may be different document sets or may be the same document set. For example, the first set of documents may be a set including all collectable documents, and the second set of documents may be a set including only documents in which a specific field of interest to provide item recommendation service is used as a major language unit. have. The controller 320 may analyze collectible documents and analyze whether each document is a document used as a major language unit in a specific field of interest to provide an item recommendation service based on language units that frequently appear.

According to another embodiment, both the first set of documents and the second set of documents may be a set including all collectable related documents. According to another embodiment, the first set of documents may be a set including all of the collectable related documents, and the second set of documents may be a set including only documents related to a specific field of interest for which an item recommendation service is to be provided.

According to another embodiment, the second set of documents may be a set including all of the collectable related documents, and the first set of documents may be a set including only documents related to a specific field of interest for which an item recommendation service is to be provided.

For step S620, the control unit 320, for example, to generate a set including only documents related to a specific field of interest to provide an item recommendation service, documents including a field language unit representing the field of interest itself, and /Or collects documents containing object language units belonging to the relevant interest field, and extracts documents whose weight of the field language unit/object language unit is greater than or equal to a preset value to generate a set containing only documents related to a specific interest field can do. The weight of the field language unit/object language unit can be determined based on the frequency, location, and context of the field language unit/object language unit. For example, documents in which the field language unit/object language unit frequently appears, the field language unit/object language unit is used as the title of the document, or displayed in large letters or fonts for emphasis are classified as documents related to a specific field of interest. You can do it.

In step S620, the controller 320 analyzes at least a part of the second set of documents, extracts a preset number of sub-language units having the highest correlation with each representative attribute language unit, and extracts two associated with each representative attribute language unit. The above sub-language units can be extracted.

Here, the controller 320 may determine a degree of association between the representative attribute language unit and the lower language unit in consideration of the frequency at which the lower language unit appears in the same or similar context as the representative attribute language unit, for example. For example, it can be seen that words appearing around language unit A in a specific sentence may appear around words associated with language unit A in other documents.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on a trip with a big heart, but because it was July, the weather was too humid, so I suffered."

Looking at the above two sentences, the word "because it's hot" has been replaced by the word "because it's wet" in the same context. The controller 320 may infer that “hot” and “wet” are words related to each other.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on vacation with a big heart, but because it was July, it was too hot and I suffered."

Similarly, the control unit 320 may infer that “travel” and “vacation” are related words through the above two sentences.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on a trip with a big heart, but because it was August, the weather was too hot and I suffered."

Similarly, the controller 320 may infer that “July” and “August” are related words through the above two sentences.

Therefore, the control unit 320 is the words "hot" and "wet" are related to each other through the previously collected documents, "July" and "August" are related words, and "travel" and "vacation" are related to each other. It can be stored as a word. It is assumed that the following sentences are collected.

"I went on vacation with a big heart, but because it was July, the weather was too humid, so I suffered."

"I went on a trip with a big heart, but it was hard because the weather was too hot because it was August."

The above two sentences are not the same context, but if you know that "hot" and "wet" are related words, "July" and "August" are related words, and "travel" and "vacation" are related words, the control unit (320) will be able to learn that "I suffered" and "I was difficult" are also related words through the above sentence.

Language unit pairs with a high frequency appearing in the same/similar context may be determined to have a high correlation. In addition, the higher the similarity between the context in which the two language units appear, the higher the correlation between the two language units can be determined. The controller 320 performs learning using the collected documents to set the degree of association between the language units, and sets the degree of association between the language units appearing in the corresponding sentence using the relationship between the set language units and the context of the sentence. In this way, it is possible to increase the accuracy of determining the association between language units.

In a similar learning method, NNLM (Neural Net Language Model),

RNNLM (Recurrent Neural Net Language Model), word2vec, skipgram, and

CBOW (Continuous Bag-of-Words) method is known. In particular, when word2vec is used, word2vec learns using documents and associates each language unit with a vector, and the similarity between the two language units can be determined by calculating the cosine similarity of the two vectors.

In this way or a similar method, the controller 320 may analyze at least a part of the second set of documents and extract a preset number of lower language units having the highest correlation with each representative attribute language unit.

In step S630, the controller 320 may extract an association weight corresponding to each pair of the representative attribute language unit and the lower language unit in the representative attribute language unit candidate set from the second set of documents.

In addition, step S630 is as follows with reference to FIG. 8 showing a detailed sequence of step S630 of FIG. 6.

Referring to FIG. 8, in step S810, at least a part of the second set of documents may be analyzed by the controller 320 to extract a degree of correlation between lower language units. For example, it is assumed that 50 sub-language units of B1 ₁ to B1 ₅₀ are collected as related sub-language units for the representative attribute language unit A1. In this case, the controller 320 may extract a degree of correlation between the two lower language units using the frequency at which the two lower language units appear together in the same document for the 50 lower language units. B1 is also interconnected between the _first B1 and B1 ₂ and B1 is determined in accordance with the _first frequency to the _second appearing in the same document. According to another embodiment, in addition to the frequency in which B1 ₁ and B1 ₂ appear in the same document affects the degree of correlation, when B1 ₁ and B1 ₂ appear in the same document, between two language units B1 ₁ and B1 ₂ The closer the distance (or the distance between sentences in which two language units appear), the higher the degree of correlation can be recognized. In a similar way, the degree of correlation between lower language units can be extracted. The distance between language units or between a language unit and a sentence is, for example, the number of sentences placed between two language units or a language unit and a sentence, the number of words placed between two language units or a language unit and a sentence, two languages. It may be determined based on one or more of the number of words positioned between a unit or a language unit and a sentence, two language units, or the number of letters between a language unit and a sentence.

In step S820, the control unit 320 may extract an association weight between each representative attribute language unit and the lower language unit based on the degree of correlation between the lower language units.

Here, the control unit 320, for example, with respect to a set of sub-language units corresponding to each representative attribute language unit, a degree of correlation between a specific sub-language unit in the sub-language unit set and another sub-language unit in the sub-language unit set, The association weight between the specific lower language unit and the representative attribute language unit may be set so that the association weight between the specific lower language unit and the representative attribute language unit has a positive correlation with each other.

For example, the higher the correlation between the sub-language units B1 ₁ of the representative attribute language unit A1 and the other sub-language units (B1 ₂ to B1 ₅₀ ) of A1, the higher the association weight between A1 and B1 ₁ may be. have. For example, an arithmetic mean (or sum) of correlations between B1 ₁ and other sub-language units (B1 ₂ to B1 ₅₀ ) of A1 may be an association weight between B1 ₁ and A1. Instead of a simple arithmetic mean, a geometric mean/harmonized mean may be used. Of the correlations between the sublingual units B1 ₁ and the other sublingual units of A1 (B1 ₂ to B1 ₅₀ ), the cutting average to be averaged excluding the highest two (example) and the lowest two (example) is It can also be used. A median may be used instead of the arithmetic mean of the correlation.

According to some embodiments, the sentence "B1 ₁ and frequency B1 of ₂ appears on the same document" is simply (or, such that appeared as a B1 ₁ and B1 ₂ is used to calculate the associated weight of B1 ₁ for A1 varies, not according to the number of which, or close to the appearance appearance) document, B1 ₁ and B1 ₂ is to appear (or that appear in the same sentence, or close to that) that B1 ₁ appeared the number of documents emerge as It can be obtained by dividing the number of documents and/or the number of documents in which B1 ₂ appears. "B1 ₁ and B1 ₂ on the frequency of appearance in such documents" in a manner similar to the B1 ₁ and B1 ₂ is an appearance number of the document and the amount of correlation (or, or close-up appearance which by appearing in the same sentence) that, as It may be set to have a negative correlation with the number of documents in which B1 ₁ appears and/or the number of documents in which B1 ₂ appears. It is simply a kind of normalization to prevent a commonly used word from having a high association weight in the representative attribute language unit A1.

Returning to FIG. 6, in step S640, a lower association degree between an object item and a lower language unit may be extracted from the first set of documents by the controller 320.

Here, among the first set of documents, a document such as an object language unit representing an object item (for example, "Taylor Swivel"), or sub-language units frequently appearing in the same sentence or adjacent sentence are determined to be related to the object item. I can.

Also, the controller 320 may collect documents in which the object language unit of the corresponding object item appears, and extract a lower association degree between the lower language unit and the object language unit according to the frequency of appearances together in the documents. In particular, when the lower language unit appears in the same sentence as the object language unit, the control unit 320 may set the association between the lower language unit and the object item to be higher than when the lower language unit appears in a sentence different from the object language unit. have.

Also, as the sentence in which the lower language unit appears is closer to the sentence in which the object language unit appears, the controller 320 may set a higher degree of association between the lower language unit and the object item of the corresponding object language unit. The proximity of two sentences is, for example, the number of sentences between two sentences, the number of words between two sentences, the number of words between two sentences, and the number of letters between two sentences. It may be determined based on more than one.

In addition, the controller 320 may set a higher degree of association between the lower language unit and the object item of the corresponding object language unit as the location where the lower language unit appears is closer to the location where the object language unit appears. The proximity between the lower language unit and the object language unit is, for example, the number of sentences positioned between the lower language unit and the object language unit, the number of words positioned between the lower language unit and the object language unit, and the lower language unit and It may be determined based on one or more of the number of words located between the object language units and the number of characters located between the lower language unit and the object language unit.

In step S650, an object-language unit association degree between the object item and each representative attribute language unit may be extracted by using the association weight of step S630 and the sub-relationship degree of step S640 by the controller 320.

For example, the degree of object-language unit association between the object item C and the representative attribute language unit A1 is the degree of sub-relation between the sub-language units of C and A1 (for example, B1 ₁ to B1 ₅₀ ) and each sub-language unit. Can be extracted using the associated weights of For example, the object-language unit association degree between the object item C and the representative attribute language unit A1 may be set higher as the lower degree of association between the object item C and B1 ₁ to B1 ₅₀ is higher.

In addition, if the sub-language unit with a higher association weight has a high sub-relationship with the object item C in the relationship with A1, the representative attribute with object C is higher than the sub-language unit with a lower association weight. The object-language unit association between language units A1 may be set to be higher. For example, in the case of Table ₁ , the lower association of the one with higher association weight (B1 ₁ ) than in the case of Table 2 is higher, so that in Table 1 is the object between object C and the representative attribute language unit A1 than in Table 2 -Language unit correlation can be set high.

Associated weight with A1 Degree of association with C B1 ₁ 0.5 0.5 B1 ₂ 0.2 0.2

Associated weight with A1 Degree of association with C B1 ₁ 0.2 0.5 B1 ₂ 0.5 0.2

According to an embodiment, the object-language unit association degree between object C and the representative attribute language unit A1 is calculated by the sum of the product of the association weight and the sub association degree corresponding to each sub-language unit (or using this total). May be. In the case of Table 1, 0.5×0.5+0.2×0.2=0.29, and in the case of Table 2, 0.2×0.5+0.5×0.2=0.20, so that the case of Table 1 is the object C and the representative attribute language unit A1 than the case of Table 2. The relationship between object-language units may be set to be high. The above-described object-language unit association calculation method is only exemplary, and the association weight with A1 in step S630 and the sub-relationship with C in step S640 are the object-language unit association and positive correlation between C and A1. If it is, other methods may be used.

Thereafter, when the communication unit 310 receives a request for recommending an item associated with a specific representative attribute language unit, the control unit 320 provides the result item through the communication unit 310 based on the object-language unit association degree extracted in step S650. I can. For example, when receiving an item recommendation request including one of the representative attribute language units, the controller 320 retrieves the information on the object items in the order of the highest object-language unit association in the relationship with the corresponding representative attribute language unit. Can provide.

In another embodiment, when receiving an item recommendation request including two or more representative attribute language units and weights corresponding thereto, the control unit 320 performs an object with representative attribute language units included in the item recommendation request for each object item- Information on object items may be provided in the order of the highest sum (or average) of values obtained by multiplying the language unit association degree by a weight (or similarly adding a weight).

9 is a flowchart of an item recommendation process according to another embodiment of the present invention.

The embodiment of FIG. 9 further includes two steps S623 and S626 between steps S620 and S630 in addition to the same processes as the embodiment of FIG. 6. Here, redundant description is avoided, and only steps S623 and S626 are described.

In step S623, the controller 320 determines whether each of the lower language units extracted in step S620 corresponds to an emotional language (emotional language). To this end, the storage unit 330 or an external server may have a dictionary of appraisal words. The appraised word dictionary is a tool for determining whether a certain word (language unit) is an appraised word, and may hold a list of appraised words, for example. A language unit included in the list of appraised words may be regarded as an appraised word, and a language unit not included in the appraised word list may be determined to be not an appraised word. However, this judgment is based on dictionary meaning, and may not reflect the popular use of words that change with the times. Therefore, the controller 320 does not determine whether the representative attribute language unit itself is an appraised word, but determines whether to use the representative attribute language unit based on whether or not the appraised words of lower language units related to the representative attribute language unit are used.

In another embodiment, the controller 320 may add another word registered as an appraised word in the appraised word dictionary and learned to have a high correlation (with a pre-set value or more) with a predetermined number or more of words to the appraised word dictionary.

In step S626, only a preset number of representative attribute language units may remain in the representative attribute language unit candidate set in the order in which the emotional language ratio (or number) of the lower language unit associated by the control unit 320 is high, and the remainder may be removed. Through this process, it is possible to prevent a language unit far from the emotional language from being treated like an emotional language.

Returning to FIG. 5, in step S520, the control unit 320 stores a basic language unit-language unit association diagram corresponding to the language unit and the representative attribute language unit pair in the storage unit 330.

Here, the language unit may include expressions that can be expressed by weights of representative attribute language units. For example, "bored" can be a language unit, and "pretty" can be a language unit.

The representative attribute language unit with a high relationship between the basic language unit and the language unit for the language unit "boring" is a representative attribute language unit capable of solving a boring situation, such as "interesting", "interesting", "for getting time". Can include.

Regarding the language unit "pretty", the default language unit-language unit is a representative attribute with a high correlation. The language unit is similar to "pretty" or "pretty", for example "beautiful", "cute", "eye-catching". It may include representative attribute language units to be described.

The process of step S520 may be performed, for example, as an input of an administrator or may be performed by receiving a basic language unit-language unit association degree determined in an external system through a network or a storage medium.

According to another embodiment, the process of step S520 may be performed in a manner similar to the process of FIGS. 6 to 9 by analyzing collectible documents such as the Internet, SNS, and news.

In addition, the process of step S520 may include a process of reflecting the user's feedback, as described later.

The process of step S520 may be performed in a manner described later with reference to any one of FIGS. 15 to 17.

11 is an example of a basic language unit-language unit association diagram according to an embodiment of the present invention.

In the embodiment of FIG. 11, all of the language units are q (C ₁ to C _q ), and all of the representative attribute language units are n (k ₁ to k _n ).

For example, the basic language unit-language unit association of the language unit C ₅ and the representative attribute language unit k ₃ is v _3,5 .

In step S530, the communication unit 310 receives and obtains a received language unit from the terminal 200, and transmits the obtained language unit to the controller 320. Here, the reception language unit is a language unit received by the terminal 200 from a user. The terminal 200 may convert a voice input into an electrical signal (voice signal) and transmit it to the item recommendation device 300. The control unit 320 of the item recommendation device 300 may analyze the voice signal, convert it into text, and match the converted text to a language unit. The controller 320 may also analyze the voice signal to analyze the intonation, pitch, speed, breathing state, etc. of the voice and use it as context information.

According to another embodiment, the terminal 200 may convert a voice input into text and transmit the converted text to the item recommendation device 300. The terminal 200 may analyze the intonation, height, speed, breathing state, etc. of the received voice input and transmit the analysis information to the item recommendation device 300. The item recommendation device 300 may utilize the analysis information as a kind of context information.

In step S540, the control unit 320 acquires a language unit-object correlation corresponding to the target language unit and each object item pair by using the object-language unit association degree and the basic language unit-language unit association degree.

12 is a detailed flowchart of step S540 according to an embodiment of the present invention.

Referring to FIG. 12, in step S1210, the control unit 320 obtains a coordinated object-language unit association corresponding to each object item and a representative attribute language unit pair for a reception language unit.

According to an embodiment, the controller 320 corresponds to the object-language unit association degree corresponding to the object item and the representative attribute language unit pair for each object item and the representative attribute language unit pair, and the reception language unit and the representative attribute language unit pair. An adjustment object-language unit association degree corresponding to each object item and a representative attribute language unit pair for a destination language unit may be obtained by applying a basic language unit-language unit association degree.

In particular, the control unit 320 is a basic language unit corresponding to the target language unit and the representative attribute language unit pair in the object-language unit association degree corresponding to the object item and the representative attribute language unit pair for each object item and the representative attribute language unit pair. -The coordinated object-language unit association degree corresponding to each object item and the representative attribute language unit pair for the destination language unit may be obtained by using a value multiplied by the language unit association degree.

In addition, the controller 320 has an object-language unit association and a positive correlation corresponding to the object item and the representative attribute language unit pair for each object item and the representative attribute language unit pair, and the reception language unit and the representative attribute language unit pair The coordinated object-language unit association degree corresponding to each object item and representative attribute language unit pair may be set for the receiving language unit so as to have a positive correlation with the basic language unit-language unit association corresponding to.

In the present invention, values representing object-language unit association, basic language unit-language unit association, coordinated object-language unit association, basic language unit-sub-language unit association, and other association relationships are all associated with higher values. Is assumed to be a close value. In other embodiments, when the association value is small, the association is tighter, and for other association values, when the association value is larger, the association is tight, so that the positive correlation-negative correlation is appropriate. It could be used instead.

For example, controller 320 receives the language units are C ₂ be when the object item i ₄ and shorthand language unit k corresponding adjustment to the _three pairs of object-language unit Relevancy object item to obtain a i ₄ and shorthand language object corresponding to the unit k ₃ pairs Relevancy default language units corresponding to the received language unit of C ₂ and k ₃ shorthand language unit pair in w _4,3-language unit Relevancy v adjusted by applying the object _{3, 4} - It is possible to obtain the degree of association of language units.

In particular, the method of applying the degree of association may be a method of multiplying the degree of association between object-relationship and basic language unit-language unit. For example, the controller 320 receives the language units are C ₂ be when the object item i ₄ and shorthand language unit adjustment corresponding to k ₃ ssang object - to obtain a Language Unit associated object entries i ₄ and shorthand The object-relationship w _4,3 corresponding to the pair of language units k ₃ is multiplied by the base language unit-language unit association v _3,2 corresponding to the target language unit C ₂ and the representative attribute language unit k3 pair (w _{4 ,3} Хv _3,2 ) can be used to obtain the coordinated object-language unit association. In another embodiment, instead of multiplying, the control unit 320 uses a function f(w _4, which applies another calculation/use method such that the coordinated object-language unit association has a positive correlation with w _4,3 and v _{3,2) . 3,} v utilizes _3,2) adjustment objects - may also obtain the language units involved. In addition, a method of using (w _4,3 Хv _3,2 ) as a coordinated object-language unit association diagram and (w _4,3 Хv _3,2 ) were adjusted based on other factors to obtain an adjusted object-language unit association. Any method you use can be used.

를 이용하여 획득될 수 있다. f(w_4,j,v_j,2)는 객체 항목 i₄ 및 수신 언어 단위 C₂, 언어 단위 k_j에 대응하는 조정 객체-언어 단위 연관도이다.In step S1220, a language unit-object association may be obtained by using a value obtained by accumulating the coordinated object-language unit association for a specific object item by the controller 320. For example, the control unit 320 sets the language unit-object association degree corresponding to the receiving language unit and the specific object item pair so as to have a positive correlation with the accumulated value of the coordinated object-language unit association degree for a specific object item. I can. The language unit-object association corresponding to the object item i ₄ and the destination language unit C ₂ pair is, for example,

It can be obtained using f(w _4,j ,v _j,2 ) is a coordinated object-language unit association diagram corresponding to the object item i ₄ , the destination language unit C ₂ , and the language unit k _j .

이 될 수 있다. 다른 예에서 객체 항목 i₄ 및 수신 언어 단위 C₂ 쌍에 대응하는 언어 단위-객체 연관도는

에 대해 기타 요인에 의한 보정을 가한 값이 될 수 있다.For example, the language unit-object association corresponding to the pair of object item i ₄ and destination language unit C ₂ is

Can be In another example, the language unit-object association corresponding to the pair of object item i ₄ and destination language unit C ₂ is

It may be a value that has been corrected by other factors.

Returning to FIG. 5, in step S550, an object item to be recommended may be provided by the control unit 320 according to a language unit-object relationship corresponding to the received language unit. For example, if the language unit-object correlation corresponding to the reception language unit C ₂ is as shown in Table 3 below, the controller 320 may provide object items to be recommended in the order of Table 4.

Object item Language unit-object association with the receiving language unit i ₁ 0.23 i ₂ 0.33 i ₃ 0.99 i ₄ 0.84

order Object Language unit-object association with the receiving language unit One i ₁ 0.99 2 i ₂ 0.84 3 i ₃ 0.33 4 i ₄ 0.23

That is, the control unit 320 may provide object items to be recommended in the order of a language unit-object association corresponding to a reception language unit in a high order. The terminal 200 receiving the object item may provide information on the object item i ₃ to the user through the display unit 220. The terminal 200 may also provide information on other object items in a lower order as necessary. The terminal 200 may provide information on the object item i ₃ to a user through a voice through a speaker instead of the display unit 220.

13 is a flowchart of an item recommendation process according to another embodiment of the present invention.

The processes of FIGS. 13 to 17 may be performed by partially using the processes of FIGS. 5 to 12 or partially changing the processes of FIGS. 5 to 12. While describing the process of FIGS. 13 to 17, descriptions of FIGS. 5 to 12 may be used if necessary.

Referring to FIG. 13, in step S1310, a representative attribute language unit candidate set is extracted from the first set of documents by the control unit 320. The controller 320 may collect, for example, language units frequently appearing in a document corresponding to the field of interest among the first set of documents as a representative attribute language unit candidate set. The process of step S1310 may be the same or similar to the process of step S610 of FIG. 6, for example. The process of step S1310 may be performed through the same or similar process as that of FIG. 7. The description of the process of FIG. 7 is not repeated again.

In step S1320, a language unit set is set by the controller 320. For example, a set of language units can be set by input from the administrator. According to a modified example, the control unit 320 may provide an interface for setting words/phrases suitable for each language unit as language unit candidates, and setting one or more of the language unit candidates as language units.

In step S1330, an object-language unit association diagram corresponding to an object item and a representative attribute language unit pair is stored by the controller 320.

The process of step S1330 may be performed according to, for example, some of the embodiments of FIGS. 6 to 8, a process similar thereto, or a process corresponding thereto. According to another embodiment, the process of step S1330 may be performed as an input of an administrator or may be performed by receiving an object-language unit association degree determined in an external system through a network or a storage medium.

Since the embodiments of FIGS. 6 to 8 have been described above, repeated descriptions are omitted. However, the process performed in step S610 of FIGS. 6 and 8 is substantially the same as the process of step S1310 of FIG. 13. Therefore, in performing the process of step S1330, even if the embodiments of FIGS. 6 to 8 are used, the process of step S610 is not performed again and the result of step S1310 may be recycled.

In step S1340, the controller 320 represents the language unit using the association weight corresponding to the representative attribute language unit and the lower language unit pair, and the basic language unit-lower language unit association degree corresponding to the language unit and lower language unit pair. A basic language unit-language unit association diagram corresponding to the attribute language unit pair is stored in the storage unit 330. The process of step S1340 may be performed, for example, as an input from an administrator, or may be performed according to one or more embodiments of FIGS. 15 to 17.

Prior to the process of step S1340 or during the process of step S1340, the lower language unit is determined, the association weight corresponding to the representative attribute language unit and the lower language unit pair is determined, and the base language corresponding to the language unit and the lower language unit pair Unit-sub-language unit association should be determined.

The lower language unit used in the process of step S1340 may be determined while step S620 is performed during the process of step S1330. In this case, the lower language unit of step S620 may be used in step S1340. When the lower language unit is not determined in step S1330, the lower language unit may be determined through the same or similar process as in step S620 of FIG. 6 and the previous process.

The association weight used in the process of step S1340 may be determined while step S630 is performed during the process of step S1330. In this case, the association weight of step S630 may be used in step S1340. When the association weight is not determined in step S1330, the association weight may be determined through the same or similar process as in step S630 of FIG. 6 and the previous process.

The degree of association between the basic language unit and the sub-language unit may be calculated, for example, in consideration of the frequency at which the language unit and the sub-language unit appear in the same or similar context.

Hereinafter, in the description of FIGS. 15 to 17, an example of obtaining a basic language unit-language unit association degree v ₃ and ₂ between the language unit C ₂ and the representative attribute language unit k ₃ will be described. For example, assume that the lower language units of the representative attribute language unit k ₃ are B ₃₁ to B ₃₅₀ . For the language unit, the representative attribute language unit, and the basic language unit-language unit association diagram, refer to the example described above with reference to FIG. 11. The relationship between the basic language unit and the lower language unit corresponding to the language unit C _j and the lower language unit Bg _h pair is expressed as x _j,h . The association weight corresponding to the pair of the lower language unit Bg _h and the representative attribute language unit k _g is expressed as y _g,h . Language unit C _j, k _g shorthand language units the language unit, and sub-adjustment corresponding to the combination of the language unit Bg _h - Fig sublanguage associated unit is expressed as x _{j, g, h.}

15 is a detailed flowchart of step S1340 of FIG. 13 according to an embodiment of the present invention.

Referring to FIG. 15, in step S1510, the control unit 320 applies an association weight to the relationship between the basic language unit and the sub-language unit to obtain an adjusted language unit-sub-language unit association.

In step S1510, the control unit 320 corresponds to the lower language unit and the representative attribute language unit pair to the basic language unit-lower language unit association degree corresponding to the language unit and lower language unit pair for each language unit and lower language unit pair. By applying the corresponding association weight, a degree of association between a sub-language unit and an adjusted language unit corresponding to each language unit and lower language unit pair for the representative attribute language unit may be obtained.

For example, controller 320 is C ₂ to the language units corresponding to the sub-language unit B _34, and shorthand language unit k ₃ pairs when C ₂ il-B3 ₄ between adjustment language unit - to obtain a degree sublanguage unit associated C ₂ unit of language and sub language unit B3 ₄ basic linguistic units corresponding to a pair-associated sub-units of language also associated weights corresponding to the sub-language unit B3 ₄ and shorthand language unit k ₃ pair in x _2,4 y _3,4 By applying, it is possible to obtain the degree of association of the sub-language unit x _2,3,4 .

In particular, the method of applying the association weight may be a method of multiplying the association weight y _3,4 corresponding to the pair of the basic language unit-sub language unit association x _2,4 and the lower language unit B3 ₄ and the representative attribute language unit k ₃ pair. . For example, when the language unit is C ₂ , the control unit 320 coordinates language unit-sub-language unit correlation between lower language unit B3 ₄ and C ₂ -B3 ₄ corresponding to the representative attribute language unit k ₃ pair x _{2,3 In} order to obtain _,4 , the base language unit-sub-language unit association x _2,4 corresponding to the language unit C ₂ and the sub-language unit B3 ₄ pairs, corresponding to the sub-language unit B3 ₄ and the representative attribute language unit k ₃ pairs. The coordination language unit-sub-language unit association degree x _2,3,4 may be obtained by using (x _2,4 Хy _3,4 ) that is the product of the association weight y _3,4 . In another embodiment, instead of multiplying, the control unit 320 is a different operation/use method in which the coordination language unit-sub-language unit association x _2,3,4 has a positive correlation with x _2,4 and y _3,4. The coordination language unit-sub-language unit association degree x _2,3,4 can also be obtained by using the function f(x _2,4 ,y _3,4 ) to which is applied. In addition, by using (x _2,4 Хy _3,4 ) as the coordination language unit-sub-language unit association x _2,3,4, and applying corrections based on other factors to (x _2,4 Хy _3,4 ) As the coordination language unit-sub-language unit association x _2,3,4, both methods can be utilized.

가 될 수 있다. 즉, 하위 언어 단위 B₃₁ 내지 B₃₅₀에 대해 언어 단위 C₂와의 기본 언어 단위-하위 언어 단위 연관도 x_2,f를 구한 뒤 각각에 대해 해당 하위 언어 단위에 대한 연관 가중치 y_3,f를 반영하여 x_2,3,f를 구하고 x_2,3,f를 누적하여 언어 단위 C₂ 및 대표 속성 언어 단위 k₃ 간의 기본 언어 단위-언어 단위 연관도를 구할 수 있다. 다른 실시예에 따르면, 언어 단위 C₂ 및 대표 속성 언어 단위 k₃ 간의 기본 언어 단위-언어 단위 연관도는

에 기타 요인에 의한 보정을 가한 값이 될 수 있다. 다른 실시예에 따르면, 언어 단위 C₂ 및 대표 속성 언어 단위 k₃ 간의 기본 언어 단위-언어 단위 연관도는

과 양의 상관 관계를 가지는 값이 될 수 있다. 여기서는 하나의 대표 속성 언어 단위에 연결돼 있는 하위 언어 단위가 50개인 것으로 가정하였으나, 대표 속성 언어 단위에 연결돼 있는 하위 언어 단위의 개수가 달라지면 수식에서 f의 누적 범위가 50이 아닌 다른 값이 될 수 있을 것이다.In step S1520, the basic language unit-language unit association between the language unit C ₂ and the representative attribute language unit k ₃ using the accumulated value of the language unit-lower language unit association x _2,3,f adjusted by the control unit 320 Can be set. That is, the basic language unit-language unit association between the language unit C ₂ and the representative attribute language unit k ₃ is

Can be. That is, for the lower language units B ₃₁ to B ₃₅₀ , the basic language unit-sub language unit association x _2,f with the language unit C ₂ is calculated, and the associated weight y _3,f for the corresponding lower language unit is reflected for each. As a result, x _2,3,f is obtained, and x _2,3,f is accumulated to obtain the basic language unit-language unit association between the language unit C ₂ and the representative attribute language unit k ₃ . According to another embodiment, a basic language unit-language unit association degree between the language unit C ₂ and the representative attribute language unit k ₃ is

It may be a value that has been corrected by other factors. According to another embodiment, a basic language unit-language unit association degree between the language unit C ₂ and the representative attribute language unit k ₃ is

It can be a value that has a positive correlation with. Here, it is assumed that there are 50 sub-language units connected to one representative attribute language unit, but if the number of sub-language units connected to the representative attribute language unit changes, the cumulative range of f in the formula will be a value other than 50. I will be able to.

16 is a detailed flowchart of step S1340 of FIG. 13 according to another embodiment of the present invention.

The processes of steps S1510 and S1520 of FIG. 16 are the same as those described above with reference to FIG. 15, and thus are not repeated.

Referring to FIG. 16, in step S1530, a basic language unit-language unit association degree between a pair of representative language units corresponding to a language unit among the representative language units corresponding to a specific language unit is a reference basic language unit-language. A basic language unit-language unit association degree for a representative language unit less than or equal to the unit association degree may be deleted. Representative language units corresponding to a specific language unit refer to language units for which a basic language unit-language unit association degree is set in relation to a specific language unit. The reference basic language unit-language unit association degree may be set in advance. According to another embodiment, the reference basic language unit-language unit association degree is set by using an average value of the basic language unit-language unit association corresponding to a specific language unit, or a basic language unit-language unit association corresponding to a specific language unit When they are arranged in order of size, they can be set using a basic language unit-language unit association of a specific rank. A specific value having a positive correlation with the basic language unit-language unit associations corresponding to other specific language units may be the reference basic language unit-language unit association. The reference basic language unit-language unit association degree may be set differently according to the language unit, or may be set identically for all language units. When the association between the basic language unit and the language unit is deleted, it is set that there is no association between the language unit and the representative language unit. The control unit may set the degree of association to 0, or the basic language unit-language of the pair of representative language units corresponding to the language unit in the list (which can be replaced by an array or other data structure) indicating the degree of association between the basic language unit and the language unit. It is also possible to delete the basic language unit-language unit association by deleting information on the unit association. A method of adding information indicating that the association between other basic language units and language units has been deleted (or the association has been deleted) may be used.

After the process of step S1530, the basic language unit-language unit association, which has a relatively fine degree of association, is deleted (i.e., it is set to have no association), so that an excessively complex operation is performed or an association that has no practical meaning to the user/manager Relationships can be suppressed.

17 is a detailed flowchart of step S1340 of FIG. 13 according to another embodiment of the present invention.

Since the processes of steps S1510 and S1520 of FIG. 17 are the same as those described above with reference to FIG. 15, the description will not be repeated. The process of step S1530 of FIG. 17 is the same as the process described above with reference to FIG. 16, so the description will not be repeated.

In step S1540, a basic language unit-language unit association degree remaining without being deleted by the controller 320 may be normalized. For example, in order to include the average value of the basic language unit-language unit relations stored connected to a specific representative attribute language unit within a specific range, the basic language unit-language unit relations stored connected to a specific representative attribute language unit are constant. It can be increased or decreased by multiplying the coefficient. For another example, basic language unit-language unit associations stored connected to a specific representative attribute language unit in order to include the total of the basic language unit-language unit associations stored connected to a specific representative attribute language unit within a specific range. It can be increased or decreased by multiplying by a certain factor. In other words, a high-value basic language unit-language unit association is concentrated in a specific representative attribute language unit, so even if a certain language unit is selected, only the specific representative attribute language unit is recommended/used or a specific representative attribute language unit is rarely used. Appropriate adjustments can be made to avoid the case of failure.

According to another embodiment, instead of multiplying a basic language unit-language unit association degree by a predetermined coefficient by the control unit 320 in step S1540, normalization may be performed by adding a predetermined coefficient or combining operations such as log and square. According to another embodiment, normalization may be performed by partially reducing only the basic language unit-language unit associations equal to or higher than a specific reference value or partially increasing only the basic language unit-language unit associations equal to or lower than a specific reference value.

According to another embodiment, in step S1540, normalization may be performed so that the average value (or sum) of the basic language unit-language unit associations stored by being connected to a specific language unit by the controller 320 is included in a specific range.

In the process of FIG. 17, step S1530 is omitted, and a modified example in which the basic language unit-language unit association normalization process is performed is also possible.

In step S1350, a reception language unit is received from the terminal 200 by the communication unit 310, and the obtained reception language unit is transmitted to the controller 320.

The reception language unit is a language unit received by the terminal 200 from a user. The terminal 200 may convert a voice input into an electrical signal (voice signal) and transmit it to the item recommendation device 300. The control unit 320 of the item recommendation device 300 may analyze the voice signal, convert it into text, and match the converted text to a language unit. The controller 320 may also analyze the voice signal to analyze the intonation, pitch, speed, breathing state, etc. of the voice and use it as context information.

According to another embodiment, the terminal 200 may convert a voice input into text and transmit the converted text to the item recommendation device 300. The terminal 200 may analyze the intonation, height, speed, breathing state, etc. of the received voice input and transmit the analysis information to the item recommendation device 300. The item recommendation device 300 may utilize the analysis information as a kind of context information.

In step S1360, the control unit 320 obtains a language unit-object association corresponding to the target language unit and each object item pair by using the object-language unit association degree and the basic language unit-language unit association degree. The process of step S1360 may be performed according to the method of step S540 of FIG. 5 or the method of FIG. 12. Description of the same content will be omitted.

In step S1370, the control unit 320 may provide an object item to be recommended according to a language unit-object relationship corresponding to the received language unit. The process of step S1370 may be performed in the same manner as the process of step S550. The same description is omitted.

14 is a flowchart of an item recommendation process according to another embodiment of the present invention.

Steps S1310, S1320, S1330, S1340, S1350, S1360, and S1370 of FIG. 14 are the same as steps S1310, S1320, S1330, S1340, S1350, S1360, and S1370 of FIG. 13, so the same description will not be repeated.

Step S1333 added in FIG. 14 may be performed at any time after the representative attribute language unit and the lower language unit are determined. For example, step S1333 may be performed simultaneously/in parallel with step S1330, or the process of step S1333 may be performed during step S1330.

In step S1333, an association weight between the representative attribute language unit and the lower language unit is stored by the controller 320. When the process of setting the association weight is not performed in step S1330, the association weight between the representative attribute language unit and the lower language unit may be set through a process similar to that of step S630 of FIG. 6. According to another embodiment, the association weight between the representative attribute language unit and the lower language unit may be stored in a manner in which the association weight set in step S1330 is called by the controller 320.

In step S1337, the control unit 320 obtains a basic language unit-lower language unit association between the lower language unit and the language unit.

The controller 320 may determine, for example, a degree of association between the language unit and the lower language unit in consideration of the frequency at which the lower language unit appears in the same or similar context as the language unit.

For example, it can be seen that words appearing around language unit A in a specific sentence may appear around words associated with language unit A in other documents.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on a trip with a big heart, but because it was July, the weather was too humid, so I suffered."

Looking at the above two sentences, the word "because it's hot" has been replaced by the word "because it's wet" in the same context. The controller 320 may infer that “hot” and “wet” are words related to each other.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on vacation with a big heart, but because it was July, it was too hot and I suffered."

Similarly, the control unit 320 may infer that “travel” and “vacation” are related words through the above two sentences.

"I went on a trip with a big heart, but because it was July, the weather was too hot and I suffered."

"I went on a trip with a big heart, but because it was August, the weather was too hot and I suffered."

Similarly, the controller 320 may infer that “July” and “August” are related words through the above two sentences.

The control unit 320 indicates that "hot" and "wet" are related words, "July" and "August" are related words, and "travel" and "vacation" are related words through the previously collected documents. It can be saved. Assume that the following sentences are collected.

"I went on vacation with a big heart, but because it was July, the weather was too humid, so I suffered."

"I went on a trip with a big heart, but it was hard because the weather was too hot because it was August."

If the two sentences are not the same context, but know that hot" and "wet" are related words, "July" and "August" are related words, and "travel" and "vacation" are related words, the controller 320 ) Through the above sentence, you can learn that "I suffered" and "I had a hard time" are also related words.

Language unit pairs with a high frequency appearing in the same/similar context may be determined to have a high correlation. In addition, the higher the similarity between the context in which the two language units appear, the higher the correlation between the two language units can be determined. The controller 320 performs learning using the collected documents to set the degree of association between the language units, and sets the degree of association between the language units appearing in the corresponding sentence using the relationship between the set language units and the context of the sentence. In this way, it is possible to increase the accuracy of determining the association between language units.

In a similar learning method, NNLM (Neural Net Language Model),

RNNLM (Recurrent Neural Net Language Model), word2vec, skipgram, and

CBOW (Continuous Bag-of-Words) method is known. In particular, when word2vec is used, word2vec learns using documents and associates each language unit with a vector, and the similarity between the two language units can be determined by calculating the cosine similarity of the two vectors.

18 is a diagram showing a term hierarchy according to an embodiment of the present invention.

When the process of FIG. 13 or 14 is completed, the hierarchical relationship between language units (C ₁ to C _q ), representative attribute language units (k ₁ to k _n ), and lower language units (BX ₁ to BX _n ) Is set.

A basic language unit-language unit association degree is set between the language unit and the representative attribute language unit, and an association weight is set between the representative attribute language unit and the lower language unit. Using this hierarchical relationship, the controller 320 may perform an operation such as recommending an appropriate object according to the language unit or selecting a new language unit candidate.

19 is a flowchart of a process of installing a virtual input interface program according to an embodiment of the present invention.

In step S1910, the terminal 200 transmits a package request message to a server providing a pressurized input interface program (hereinafter, a program providing server) (not shown). The package request message is a message requesting a virtual input interface program. In step S1920, the program providing server delivers a virtual input interface program (package) to the terminal 200.

In step S1930, the terminal 200 installs the received virtual input interface program (package). If installation is unnecessary and the received virtual input interface program can be executed as it is, step S1930 may be omitted. The terminal 200 executes a virtual input interface program according to a user's input to perform a method of providing item information distribution according to an event and a method of recommending an item according to an event.

20 is a flowchart illustrating a method of setting a degree of association and order between a language unit and a use item according to an embodiment of the present invention.

A method of setting a degree of association and an order between a language unit and a usage item according to an embodiment of the present specification is a method of providing information on a search target matched according to a language unit selected by a user through a category. The language unit may be a word, a phrase, or a sentence, and includes an appraisal word representing tastes, preferences, preferences, and emotions of the user.

2, 3, and 20, in step S2011, the item recommendation device 300 may classify an image of an object into a plurality of categories. The category may classify objects with strong user preferences according to colors, such as artwork, products, contents, places, and food, according to their use. The product may include fashion products including clothes, bags, wallets, and fashion accessories, and beauty products including cosmetics and hair. Content may include movies, broadcasts, and art. Places may include accommodation, activities, attractions, and cities. Food may include dishes and restaurants.

Through the classification of categories, it is possible to filter only images of objects for the category desired by the user. To this end, the item recommendation device 300 guides the user to select a filtering condition, and extracts only a document written for an image corresponding to the filtering condition selected by the user. Here, the filtering condition may include the category. That is, the item recommendation device 300 may extract and analyze only documents of a category selected by the user from among the acquired documents. For example, when an art work is selected by the user, only a document related to the art work may be extracted, and the following steps may be performed on the extracted document. All extracted documents may be evaluated as having equal value, or newer documents may be evaluated as having higher value.

In addition, when a plurality of language units appear in one document, only a representative language unit may be selected based on the frequency of appearance of the plurality of language units. Alternatively, a plurality of language unit pairs may be selected. Depending on the frequency of appearance of the language unit, the weight may be different when calculating the relationship later. The higher the frequency of appearance of language units, the higher the degree of association or the greater the weight. That is, the correlation between the frequency of the language unit and the image of the document has a positive correlation.

In step S2012, the item recommendation device 300 collects a document created in association with the image and provides a plurality of classified categories from the collected document to the terminal 200.

Documents include comments or posts related to the image. For example, if there is a text that appears in the same document as the image, the language unit and phrase of the text can be treated as related to the image. According to another example, when the distance between the image and the language unit (or phrase, etc.) is less than or equal to a preset value, the corresponding language unit may be treated as related to the image. The distance between an image and a language unit can be calculated using one or more of the number of images located between the two, the number of bytes of text, the number of characters, the number of words, the number of lines, the number of words, and the number of sentences. For example, if the distance between the image and the language unit is calculated using the number of words, the distance between the image and the language unit may be a value proportional to the number of words between the image and the corresponding language unit.

For example, the item recommendation device 300 may include articles and shopping malls posted on SNS (Social Network Services) such as web pages, blogs, various cafes, community sites, Twitter or Facebook, etc. You can acquire text, comments, or comments that are left in relation to images on sites and Twitter.

In step S2013, the terminal 200 receives interface information that enables a user to select an arbitrary category from among a plurality of classified categories provided from the item recommendation device 300 and displays the category selection unit.

In step S2014, the category selected by the user 400 is acquired, and in step S2015, the acquired category is transmitted to the item recommendation device 300.

In step S2016, the item recommendation device 300 sets a degree of association between language units and items for a corresponding category. Here, the item recommendation device 300 analyzes the language units that appear together in the corresponding category. To this end, the storage unit 330 or an external server may hold a dictionary of emotion words (emotional language). The appraised word dictionary is a tool for determining whether a certain word (language unit) is an appraised word, and may hold a list of appraised words, for example. A language unit included in the list of appraised words may be regarded as an appraised word, and a language unit not included in the appraised word list may be determined to be not an appraised word. However, since such a judgment is based on dictionary meaning and may not reflect the use of words by the public that changes according to the times, appraisal words may be added or deleted by the administrator.

In step S2017, the item recommendation device 300 acquires user usage details. Here, the user usage details may be items used by the user, for example. The items used by the user may include various items that the user interacts with, such as items purchased by the user, items inquired, items selected, and items that are recommended (liked). The usage history may include only a list of items used, or may further include information on the order of use. The usage history may include a list of items used and information at the time of use.

In step S2018, the item recommendation device 300 provides a related language unit-related degree list corresponding to the usage details as an input of the machine learning system. The machine learning system may form a part of the item recommendation device 300 or may exist as an entity outside the item recommendation device.

The following shows an example of the input of a machine learning system.

[Time 1 (Sequence 1), [Linked Language Unit, Association] List]

[Time 2 (Sequence 2), [Linked Language Unit, Association] List]

·

·

·

[Time n (Sequence n ), [Linked Language Unit, Association] List]

In step S2019, the item recommendation device 300 obtains a recommended language unit-relevance list by inputting language unit-relevance lists corresponding to the usage details into the machine learning system.

Here, in the machine learning system, the list of the next nth [linked language unit, degree of relevance] pair is affected according to the list of [associated language units, degrees of association] pairs in the previous 1st to n-1th order (time). learning to have received at home, and the (n + 1) th [associate language units, the association also] a list of pairs of 1 ~ n-th [associate language units, the association also] be a machine learning system that can be recommended based on the list of pairs have.

Machine learning systems that can operate in this way are, for example, some of Tensor Flow, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) (reference paper http://arxiv.org/abs/1808.03314), Or a combination of some of these.

In step S2020, the item recommendation device 300 provides the recommended items to the terminal 200 in an order corresponding to the recommended language unit-relevance list.

Also, if the degree of association between the language unit i and the usage history (user) is R _i and the degree of association between the specific item I _j and the language unit i is R _ij , the degree of association between the specific item I _j and the usage history (user) is Σ _It can be expressed as _i=0...k- f(R _ij , R _i ). Here, k is the number of language units selected in step 1).

In addition, f(x, y) may be, for example , a product of x and y, or may be another operation that allows x and y and f(x, y) to have a positive correlation. Other operations that have a positive correlation with each f(x, y) value other than f(x, y) accumulation may be used. In the modified example, a specific item-use history (user) association and _i=0...k-1 f(R _ij , R _i ) cumulative value are positive, not a simple accumulation ( _i=0...k-1 ) . Other equations, including proportional operations, logarithmic operations, etc. that allow correlation, can be used to find the item-user association.

In step S2021, the terminal 200 displays and provides the recommended item to the user.

19 is an exemplary view showing main language units for each language unit list based on a language unit associated with a selected item according to an embodiment of the present invention.

The aforementioned machine learning input will be described with reference to FIG. 21.

21 is a diagram illustrating an example of a machine learning system used to set a degree of association and order between a language unit and a use item according to an embodiment of the present invention.

Referring to FIG. 21, the LSTM layer of the machine learning system used to set the degree of association and order between language units and usage items according to an embodiment of the present invention is c (cell state) called Input Gate, Forget Gate, and Output Gate. There are gates that control the value of. Each gate controls its value through the sigmoid layer and pointwise multiplication. Each gate serves to protect and control the cell state.

Here, the input gate layer is a gate that determines which of the new information is to be contained in the cell state. Next, the tanh layer generates new data candidate values, and prepares to add the generated candidate values to the cell state.

The Forget gate layer is a gate that actually forgets what you choose to forget, using weights for how much to forget. Update the state by multiplying the previous state c by the output value of the forget gate. Also, the output values from the input gate are multiplied and added to the state.

The output gate layer is a gate that determines what value to output.

Here, it is the sigmoid layer of the gate that determines which part of the cell state is output. The final output value is calculated by multiplying the value between -1 and 1 generated by tanh the cell state by the sigmoid output value.

22 is an exemplary diagram illustrating main language units for each language unit list based on language units associated with a selected item according to an embodiment of the present invention.

Referring to FIG. 22, the main language units for each language unit list are analyzed with word2vec based on the language unit associated with the selected item, and the analyzed result is mapped in two dimensions using a dimension reduction method. For each language unit list (for example, user response/atmosphere/genre characteristics) on the basis of language units, the average correlation with the selected items is arranged and displayed in the highest order.

As shown in FIG. 22, for example, in the'user response' language unit list, "want to eat" has the highest average correlation with selected items, followed by "hungry", and then " It is the order of "I want to see".

In addition, in the'mood' language unit list, "cute" has the highest average correlation with the selected items, followed by "remembering", followed by "normal".

In addition, in the list of'genre characteristics' language units, "complex" has the highest average association with the selected items, followed by "fire", followed by "clean".

23 is an exemplary view showing main language units for each language unit list based on a language unit associated with a selected item according to another embodiment of the present invention.

Referring to FIG. 23, based on the language unit associated with the selected item, the main language units for each language unit list are analyzed with word2vec, and the analyzed result is mapped in two dimensions using a dimension reduction method. For each language unit list (for example, performer/atmosphere/user reaction/genre characteristics) on the basis of language units, the average correlation with the selected items is arranged-displayed in the order of the highest.

As shown in Fig. 23, for example, in the list of'performers' language units, "cute" has the highest average correlation with selected items, followed by "cute", and then "clearly" It is the order of "smile".

In addition, in the'atmosphere' language unit resource, the average correlation with the items selected for "memory" is the highest, followed by "cute", followed by "exciting".

In addition, in the'user response' language unit list, the average correlation with the items selected for "want to see" is the highest, followed by "thank you", followed by "crazy".

In addition, in the'genre characteristic' language unit list, the average association with the items selected for "suffering" is highest, followed by "complex", followed by "quality".

24 and 25 are exemplary diagrams showing an average degree of association of major language units for each list according to an embodiment of the present invention.

Referring to FIGS. 24 and 25, charts are prepared based on an average degree of association corresponding to each language unit list. That is, the average degree of association of the major language units for each list is analyzed with word2vec, and the analyzed result is mapped in two dimensions using the dimension reduction method. Here, a circle with a darker color or a larger color is a language unit having a higher relevance to the selected item, and language units having a similar position are language units having a high similarity (the position of a vector is similar).

24 and 25, the upper charts are web charts showing the degree of relevance of language units with high relevance as a graph, but are not limited thereto and need not necessarily be web charts. Also, the charts below are scatter charts, a kind of one chart.

26 is an exemplary diagram showing an average degree of association of major language units for each list according to another embodiment of the present invention.

Referring to FIG. 26, charts are prepared based on an average degree of association corresponding to a list of each language unit. That is, the average degree of association of the major language units for each list is analyzed with word2vec, and the analyzed result is mapped in two dimensions using the dimension reduction method. Here, a circle with a darker color or a larger color is a language unit having a higher relevance to the selected item, and language units having a similar position are language units having a high similarity (the position of a vector is similar).

In FIG. 26, the upper chart is a web chart showing the degree of relevance of language units with high relevance as a graph, but is not limited thereto and does not necessarily need to be a web chart. Also, the chart below is a scatter chart, which is a kind of one chart.

27 and 28 are exemplary views showing a language unit and a language unit list of main language units for each list according to an embodiment of the present invention.

In FIGS. 27 and 28, the chart is a scatter chart, which is a kind of one chart, and zoom-in and zoom-out are possible. Regardless of the zoom-in state/zoom-out state, when the mouse is turned on (otherwise, a corresponding method) on each circle, the corresponding language unit + A list of language units is displayed. FIG. 27 is an enlarged exemplary view in the vicinity of "normal" in FIG. 26; Here, a circle with a darker color or a larger color is a language unit having a higher relevance to the selected item, and language units having a similar position are language units having a high similarity (the position of a vector is similar).

29 is an exemplary view showing a language unit and a language unit list of main language units for each list according to another embodiment of the present invention.

In FIG. 29, the chart is a scatter chart, which is a kind of one chart, and zoom-in/zoom-out is possible. Regardless of the zoom-in state/zoom-out state, when the mouse is turned on (other, corresponding method) on each circle, the corresponding language unit + language unit A list is displayed. Here, a circle with a darker color or a larger color is a language unit having a higher relevance to the selected item, and language units having a similar position are language units having a high similarity (the position of a vector is similar).

Here, when the above-described web chart or one chart is clicked, as shown in FIGS. 30 and 31 to be described later, the language unit adjustment interface corresponding to the clicked list is displayed.

FIG. 30 is an example showing a language unit adjustment interface corresponding to a clicked list according to an embodiment of the present invention, and FIG. 31 is a diagram illustrating an entire language unit adjustment interface corresponding to a clicked list according to an embodiment of the present invention. It is an exemplary diagram shown.

In FIGS. 30 and 31, in the initial language unit selection, n language units having a high average correlation with the corresponding items are selected based on the selected items based on a predetermined threshold value. Here, the threshold value may be the maximum number or the minimum degree of association.

In addition, as illustrated, the language units that have already been selected are dark purple, and as the degree of association decreases, the background color becomes blurred, and are arranged in the order of the higher degree of association. In addition to the hue and shading method, it is also possible to display a lower degree of relevance by using another method, for example, brightness/saturation/color/shape.

In addition, by clicking "Select language unit" in FIG. 31, it is possible to individually add/delete the entire language unit.

FIG. 32 is an exemplary view showing a language unit adjustment interface corresponding to a clicked list according to another embodiment of the present invention, and FIG. 33 is an entire language unit adjustment interface corresponding to a clicked list according to another embodiment of the present invention. It is an exemplary diagram showing.

In FIGS. 32 and 33, in the initial language unit selection, n language units having a high average correlation with the corresponding items are selected based on the selected items based on a predetermined threshold value. Here, the threshold value may be the maximum number or the minimum degree of association.

In addition, as illustrated, the language units that have already been selected are dark purple, and as the degree of association decreases, the background color becomes blurred, and are arranged in the order of the higher degree of association. In addition to the hue and shading method, it is also possible to display a lower degree of relevance by using another method, for example, brightness/saturation/color/shape.

In addition, by clicking "Select language unit" in FIG. 33, it is possible to add/delete all language units.

The recommended items recommended by the item recommendation method using a degree of association and an order between a language unit and a usage item according to the present invention are as illustrated in FIG. 34.

As described above, the item recommendation method using the degree of association and order between the language unit and the item to be used according to the present invention uses the degree of association and the order between the language unit and the item to be used, and recommends efficient, effective and highly purchaseable information. I can.

As described above, the present invention has been described based on preferred embodiments, but these embodiments are intended to illustrate, not limit, the present invention, so those skilled in the art to which the present invention pertains without departing from the spirit of the present invention. Various changes, changes or adjustments to the example will be possible. Therefore, the scope of protection of the present invention should be construed as including all examples of changes, examples of changes or adjustments belonging to the gist of the present invention.

150: communication network 200: terminal
210: input unit 220: display unit
230, 310: communication unit 240, 330: storage unit
250, 320: control unit 300: item recommendation device

Claims (10)

An item recommendation method using a degree of association and order between a language unit performed by an item recommendation device and an item to be used,
Obtaining a pre-set language unit-item association for each category by a control unit of the item recommendation device;
Obtaining a usage history of the category by a control unit of the item recommendation device;
Inputting, by a control unit of the item recommendation device, language unit-relevance lists corresponding to the usage details into a machine learning system, and obtaining a recommended language unit-relevance list; And
Providing recommended items to the terminal in an order corresponding to the recommendation-related language unit-relevance list by the control unit of the item recommendation device; and an item recommendation method using a degree of association and order between a language unit and a used item.
The method of claim 1,
Further comprising the step of obtaining the category identification information selected by the user,
The step of obtaining the use details of the category by the control unit of the item recommendation device includes obtaining the use details of the category corresponding to the category identification information by the control unit of the item recommendation device. Item recommendation method using relevance and order.
The method of claim 2,
An item recommendation method using a degree of association and an order between a language unit and a used item, further comprising the step of providing interface information to the terminal to enable selection of an arbitrary one category from among a plurality of classified categories.
The method of claim 1,
An item recommendation method using a degree of association and order between a language unit and a used item, further comprising the step of selecting a language unit having a high correlation with the corresponding item as a language unit associated with the item using a preset threshold.
The method of claim 4,
The step of selecting a language unit having a high relevance to the item as a language unit associated with the item by using the preset threshold value,
An item recommendation method using a degree of association and order between a language unit and a used item, comprising the step of selecting a language unit having a high correlation with the corresponding item as many as a preset maximum number.
The method of claim 4,
The step of selecting a language unit having a high relevance to the item as a language unit associated with the item by using the preset threshold value,
An item recommendation method using the degree of association and order between the language unit and the used item for selecting a language unit with a degree of association with a corresponding item equal to or greater than a preset minimum degree of association.
The method of claim 1,
The step of obtaining a recommended language unit-relevance list by inputting a language unit-relevance list corresponding to the usage details into a machine learning system by the control unit of the item recommendation device,
According to the list of pairs of [associated language units, degree of association] in the 1st to n-1th order, the list of the next nth [linked language unit, degree of association] pair is assumed to be affected, and the n+1th [ A language that obtains a list of recommended language units-relevance degree using a machine learning system that can recommend a list of related language units, degree of association] based on a list of the 1st to nth pairs of [related language units, degree of association] Item recommendation method using the degree of association and order between units and items used
The method of claim 1,
Generating a user recommendation item list based on a language unit-user association degree by the control unit of the item recommendation device,
Individual values having a positive correlation to the language unit-user association for a specific language unit and the language unit-item association for the specific language unit have a positive correlation with the accumulated value for all language units Establishing an item-user association; And
An item recommendation method using a degree of association and an order between a language unit and a used item, including the step of generating a list of user recommended items according to the item-user association degree.
The method of claim 8,
Individual values having a positive correlation to the language unit-user association degree for the specific language unit and the language unit-item association degree for the specific language unit have a positive correlation with the accumulated value for all language units To set the item-user association,
Item-user correlation to have a positive correlation with the accumulated value for the entire language unit by multiplying the language unit-user association for a specific language unit and the language unit-item association for the specific language unit Item recommendation method using a degree of association and an order between a language unit and a use item including the step of setting a.
The method of claim 1,
Providing recommended items to a terminal in an order corresponding to the recommended related language unit-relevance list by the control unit of the item recommending device,
If the degree of association between the language unit i and the usage details is R _i and the degree of association between the specific item I _j and the language unit i is R _ij , the degree of association between the specific item I _j and the usage details is Σ _{i = 0...k -} f and (R _ij, R _i), where, k is the item associated with the order and also between one of the language units the language unit number and the entry in the first time using methods like.

Images