JP7209297B2 - Preference analyzer - Google Patents

Description

The present invention relates to technology for analyzing user preferences.

It is said that there are nearly 2 billion websites on the Internet. It can be said that the possibility that the desired information exists on the Internet has increased due to the abundance of information, but it can also be said that it has become difficult to find the most suitable information. Additionally, users often fail to clearly recognize or express what information they really want.

Under such circumstances, studies are actively being conducted to analyze preferences such as user preferences, interests, thoughts, beliefs, etc. from behavior histories on the Internet. Appropriate analysis of user preferences and filtering of information based on the analysis results will make it easier for users to find the information they desire. It is also believed that the accuracy of targeting advertisements can be further improved.

Japanese Patent Application Laid-Open No. 2003-141180 JP-A-9-245058

In order to deeply analyze the user's preferences (thinking tendencies), the present inventors have investigated the predispositions that form thoughts and feelings that even the user is not aware of, in other words, the "thoughts" hidden in the subconscious of the user. I think that it is necessary to draw out the “source of the essence” and verbalize it.

The present invention has been completed based on the recognition of the above problems, and its main purpose is to provide a technique for analyzing user preferences through dialogue with the user.

A preference analysis device according to an aspect of the present invention includes: a term acquiring unit that receives input of derived terms evoked from reference terms by a user to be analyzed; It includes a related setting unit that updates the term-related information to be displayed, and an important term specifying unit that specifies important terms for the user.
The relationship setting unit updates the term-related information when a new derived term is acquired after setting the derived term as the next reference term.
The important term identification unit identifies important terms from a plurality of terms based on the appearance of each of the plurality of terms included in the term-related information.

ADVANTAGE OF THE INVENTION According to this invention, it becomes easy to deeply analyze a user's taste.

FIG. 1(a) is a schematic diagram for explaining a preference analysis method in a server type. FIG. 1(b) is a schematic diagram for explaining a preference analysis method in the edge type. FIG. 4 is a schematic diagram for explaining a document information sorting method according to the embodiment; 3 is a functional block diagram of a user terminal; FIG. It is a screen figure of a dialog screen. FIG. 4 is a schematic diagram showing a process of generating term-related information; FIG. 4 is a schematic diagram showing the structure of term-related information; It is a figure which shows the evaluation coordinate of an important term. 4 is a flow chart showing a user analysis process; FIG. 4 is a diagram showing evaluation coordinates of document information; 4 is a flow chart of a document analysis process and a match determination process;

In this embodiment, a method of analyzing the user's preference (thinking tendency) through dialogue with the user and selecting content (document information) that matches the preference will be described. First, as a system for analyzing user preferences, two types of configurations, a server type and an edge type, will be described. Note that the edge type is assumed in this embodiment.

FIG. 1(a) is a schematic diagram for explaining a preference analysis method in a server type.
A plurality of companies (servicers) transmit document information from a plurality of servers 102 (own company sites). The "document information" referred to here may be any digital content including text information relating to advertisements, comments, introductions, news, and the like. In FIG. 1, company A operates server 102a and company B operates server 102b. Company A mainly extracts information such as user P1's sex, age, order history, and access time zone from the access history of user P1 to server 102a, and stores this as user data in external database 104a. Based on the user data, company A analyzes user P1's preferences, for example, information about what he/she is interested in, what he likes, and what he cares about. Company B also independently accumulates user data in the external database 104b based on the access history of the user P1 to the server 102b.

Company A selects document information that may be of interest to user P1 based on the user data accumulated in the company's external database 104a, and transmits the document information to user terminal 100. FIG. The user terminal 100 may be a mobile terminal such as a smart phone, a laptop computer, or a tablet computer. The same applies to Company B. In this manner, the current mainstream method is that the server 102 manages user data, analyzes user preferences from the user data, and selects information to be provided to the user by the server 102 .

In the case of the server type, personal information is collected by the company, so users tend to feel resistance and caution. In addition, since users access various servers 102 in addition to the server 102a, it is nearly impossible for one company to collect all user data. As a result, there is a possibility that the document information provided by the company to the user P1 and the preference of the user P1 will be different. In the case of the server type, the user often feels as if he/she is being forced to unilaterally receive uninteresting document information.

FIG. 1(b) is a schematic diagram for explaining a preference analysis method in the edge type.
In the edge type, software that functions as a concierge (hereinafter referred to as “concierge function”) is installed in the user terminal 100 . A user terminal 100 equipped with a concierge function functions as a "preference analyzer". The concierge is a so-called chatbot, and accumulates user data obtained through interaction with the user in the user terminal 100 (internal database 106). The internal database 106 is a non-volatile memory built into the user terminal 100 . The concierge function of the user terminal 100 periodically visits the server 102, downloads and verifies the document information, and determines whether or not the information should be provided to the user.

In the edge type, the concierge function analyzes user preferences based on user data accumulated in internal database 106 . Users do not need to pass personal information to the outside. Since the user data is collected by the single user terminal 100 that is the starting point of the action, rather than by a large number of servers 102 that act as counterparts, it is easier to collect and enrich the user data than in the server type, and high-precision information filtering is possible. It is possible.

FIG. 2 is a schematic diagram for explaining a document information sorting method according to the present embodiment.
The user terminal 100 (preference analysis device) analyzes the user's preferences and acquires various document information from the server 102 . The user terminal 100 presents the document information to the user if the document information matches the user's taste, in other words, if the document information is presumed to be the information that the user wants. If not, do not present. There are three processes for selecting document information: (1) a user analysis process for analyzing user preferences, (2) a document analysis process for analyzing characteristics of document information, and (3) a match determination process for determining compatibility between users and document information. Includes processing. First, an outline of the three types of processing will be described, and details will be described later with reference to FIG. 4 and subsequent figures.

(1) User Analysis Process First, the user inputs a “term” on the user terminal 100 . Although the details will be described later, the user inputs terms one after another into the user terminal 100 by using the free association method. The user terminal 100 creates "term-related information" that indicates connections between terms. This term-related information is information indicating connections between words that come to mind in the user's mind, and is, so to speak, information that simulates a "mental network." Term-related information will be described in detail with reference to FIGS. The user terminal 100 extracts "important terms" that characterize the user's preferences from the term-related information. An important term is a verbalization of a concept assumed to be an important predisposing factor (element of thought) when a user forms a preference. The user then rates the key terms. Thus, in the user analysis process, important terms are found and the user's evaluation (impression) of the important terms is set.

(2) Document Analysis Process The company provides various document information from the server 102 . The user terminal 100 acquires document information as a web page from the server 102 and extracts "feature terms" as terms that characterize the content of the document information. In other words, the user terminal 100 defines characteristics of document information from one or more feature terms.

(3) Conformity Judgment Process The user terminal 100 defines user preferences based on important terms and their evaluation values, defines characteristics of document information based on feature terms, and compares the document information and the user. Determine compatibility with For example, document information containing feature terms similar to important terms that the user has a strong interest in is judged to have high relevance. The user terminal 100 filters a large amount of document information by presenting document information that is considered to match the user's taste.

FIG. 3 is a functional block diagram of the user terminal 100. As shown in FIG.
Each component of the user terminal 100 includes computing units such as a CPU (Central Processing Unit) and various coprocessors, storage devices such as memory and storage, hardware including wired or wireless communication lines connecting them, and storage devices. It is implemented by software that is stored and provides processing instructions to the calculator. A computer program may consist of a device driver, an operating system, various application programs located in their higher layers, and a library that provides common functions to these programs. Each block described below represents a functional block rather than a hardware configuration.

User terminal 100 includes user interface processing unit 110 , communication unit 114 , data processing unit 112 and data storage unit 116 .
The user interface processing unit 110 receives operations from the user via the touch panel, and is in charge of user interface processing such as image display and audio output. The communication unit 114 is in charge of communication processing with the server 102 and other user terminals 100 via the Internet. The data storage unit 116 stores various data. The data processing unit 112 executes various processes based on data acquired by the user interface processing unit 110 and the communication unit 114 and data stored in the data storage unit 116 . Data processing unit 112 also functions as an interface for user interface processing unit 110 , communication unit 114 and data storage unit 116 .

The user interface processing unit 110 includes an input unit 118 that receives input from a user, and an output unit 120 that outputs various information such as images and sounds to the user.

The input unit 118 includes a term acquisition unit 122 and an evaluation input unit 124 .
The term acquiring unit 122 receives term input from the user. The evaluation input unit 124 receives an input of an evaluation value for important terms from the user.

Communication unit 114 includes a transmission unit 138 that transmits data and a reception unit 140 that receives data. Reception unit 140 includes document acquisition unit 142 . The document acquisition unit 142 crawls the Internet and automatically downloads web pages (document information) from various servers 102 .

The data processing unit 112 includes a relation setting unit 126 , an important term specifying unit 128 , an important term evaluating unit 130 , a characteristic term extracting unit 132 , a characteristic term evaluating unit 134 and a conformity determining unit 136 .
The relationship setting unit 126 sets relationships (links) between terms and updates term-related information (described later with reference to FIGS. 5 and 6). The term-related information is stored in data storage unit 116 . The important term identification unit 128 extracts important terms based on the term-related information (described later with reference to FIG. 6). The important term evaluation unit 130 sets the evaluation value of the important term according to the input from the user (described later with reference to FIG. 7). The feature term extraction unit 132 extracts feature terms from the document information (described later with reference to FIG. 9). The feature term evaluation unit 134 calculates the evaluation value of the feature term (described later with reference to FIG. 9). The suitability determination unit 136 determines suitability of the document information for the user (described later with reference to FIG. 9).

FIG. 4 is a screen diagram of the dialogue screen 150. As shown in FIG.
When the user activates the concierge function on the user terminal 100, the output unit 120 causes the interactive screen 150 to be displayed. A user interacts with butler 152 (a character). Through interaction between the user and butler 152, the concierge function explores the user's preferences.

On the interactive screen 150 of FIG. 4, the butler 152 asks the user, "Is there anything that interests you?" On the other hand, the user has entered "Okinawa" in the term input area 154 . When the user touches the reply button 156, "Okinawa" is registered as a term of interest to the user.

Then, butler 152 asked, "What do you think of Okinawa?" "What do you like about Okinawa?" "What do you like about Okinawa?" ” to users. For example, it is assumed that the user answers "Mimigar" to the question "What do you think of Okinawa?". In this case, the term ``Okinawa'' is the term that serves as the starting point of association (hereinafter referred to as the ``standard term''), and the term ``Mimigar'' is a term evoked from the standard term ``Okinawa'' (hereinafter referred to as the ``derivative term''). is. The relation setting unit 126 sets a relation (link) between the reference term “Okinawa” and the derivative term “Mimigar” and registers them in the term relation information.

Butler 152 then asks a similar question using "mimigar" as the new reference term. When the user answers "niku" as a term evoked from "mimi ga", the term "niku" is registered in the term-related information as a derived term of the reference term "mimi ga". In this way, by repeating the process of inputting a derived term based on a reference term and inputting the next derived term using the derived term as a new reference term, the term acquiring unit 122 acquires a plurality of terms from the user. Also, each time a new term is acquired, the association setting unit 126 updates the term association information.

FIG. 5 is a schematic diagram showing the process of generating term-related information.
It is assumed that the user recalls "mimi ga" from "Okinawa" (S1), "mimi ga" from "mimi ga" (S2), and "beer" from "meat" (S3). Further, the user recalls "whiskey" from "beer" (S4), "Ireland" from "whiskey" (S6), and "beer" from "Ireland" again (S6). Since the user recalls the term "beer" twice, it is possible that "beer" is an important keyword for this user in exploring the user's tastes, for example, what they want, what they like, and what they care about. have a nature.

After "beer", "summer" (S7), "Okinawa" (S8), and "sea" (S9) are played early. Since "Okinawa" is also a reappearing term, "Okinawa" is also considered to be a keyword that has established an important position in the subconscious of this user. In this way, terms that are likely to be recalled as a result of free association are “candidates for important terms”.

FIG. 6 is a schematic diagram showing the structure of term-related information.
As described above, users enter various terms into user terminal 100 . The relationship setting unit 126 enriches the term-related information by setting a relationship (associative relationship) between the reference term and the derivative term each time the term acquisition unit 122 acquires a term. The first reference term may be freely input by the user, or the related setting unit 126 may randomly propose a theme via the butler 152 . For example, the butler 152 may first ask, "When you think of Osaka?" In this case, the first reference term “Osaka” is initially set by butler 152 (relationship setting unit 126) instead of by the user. On the other hand, the user may freely input the first reference term such as "Okinawa" shown in the example of FIG.

The user inputs the derivative term 2 from the reference term 1 , sets the derived term 2 as a new reference term 2 , and then inputs the next derived term 3 . Also, multiple types of reference terms may be provided. For example, the term acquiring unit 122 may acquire a derived term group evoked with the reference term “Osaka” as a starting point, and then acquire a derived term group with a completely different reference term “Jupiter” as a starting point. Some terms are entered multiple times, while others are entered only once.

In this embodiment, the important term identification unit 128 selects terms that are ranked within the top 10 in terms of the number of relationships (number of links) from these term groups and that are related to five or more other terms. Identify as “key terms”. Circles in FIG. 6 indicate terms. In FIG. 6, "UK", "Rock", "Shimonoseki", "Korea", etc. are identified as important terms because they have many relationships. Important terms are terms that occupy an important position in the user's deep psychology, among the huge group of terms that the user recalls through dialogue with the butler 152, and serve as important clues in analyzing the user's preferences. should be a keyword.

FIG. 7 is a diagram showing evaluation coordinates of important terms.
A key term identifier 128 identifies one or more key terms, and an output component 120 presents the key terms to the user. The user enters rating values for key terms. For example, with respect to the key term "Britain", butler 152 asks the question "Do you like England?" and "Are you interested in England?" The user enters '5' if they love the UK, '-5' if they hate it, and '0' if they do not like or dislike it. Numerical inputs rate how much the user likes "UK". The evaluation input unit 124 acquires an evaluation value from the user with respect to the evaluation axis of "like/dislike" in the UK (hereinafter referred to as "like/dislike evaluation"). Similarly, an evaluation value is obtained for the evaluation axis of "interest" (hereinafter referred to as "interest evaluation").

The important term evaluation unit 130 similarly obtains evaluation values from the user for other important terms, thereby specifying the coordinates of the important terms in the “evaluation space”. The “evaluation space” here is a conceptual space formed by expressing evaluation values for various terms in coordinates. The evaluation space shown in FIG. 7 is a two-dimensional space composed of two axes of favorable evaluation and interest evaluation. According to FIG. 7, the user likes "golf" a little, but is not so interested in it, and does not like, dislike, or is not interested in "Shimonoseki". On the other hand, it can be seen that users not only have a strong interest in "rock" but also love it. The evaluation space schematically expresses important terms that occupy an important position in the user's deep psychology and the user's impression of the important terms.

The evaluation space need not be a two-dimensional space, and may be a high-dimensional space whose coordinates are defined by three or more evaluation axes. When the dimension is three or less, the output unit 120 may graphically display the evaluation space shown in FIG. 7, but it is not essential to graphically display the evaluation space.

FIG. 8 is a flow chart showing the user analysis process.
As described above, in the user analysis process, the butler 152 elicits terms (words) from the user by asking questions to the user, identifies important terms, and obtains the user's evaluation value for the important terms. explore your tastes.

The term acquiring unit 122 first acquires terms from the user (S10). The association setting unit 126 updates the term association information each time a new term is acquired (S12). When continuing term input (N of S14), the process returns to S10, and a derivative term recalled from the previously input term or a new reference term is input. When term input ends (Y of S14), the important term identification unit 128 identifies important terms from the obtained term group (S16). The output unit 120 presents the important terms to the user, and the user inputs an evaluation value for the important terms (S18). The important term evaluation unit 130 associates the important terms with their evaluation values and registers them in the data storage unit 116 .

FIG. 9 is a diagram showing evaluation coordinates of document information.
In FIG. 9, the like/dislike evaluation axis is the x-coordinate, and the interest evaluation axis is the y-coordinate. It is assumed that the document acquisition unit 142 acquires document information D1 related to "digital arts of hot springs in Kyushu". The user terminal 100 determines whether this document information D1 is appropriate information for the user, in other words, whether it is valuable information that should be presented to the user.

First, the feature term extraction unit 132 extracts feature terms from the document information D1. A characteristic term is a keyword that characterizes the document information D1. First, the feature term extraction unit 132 extracts terms (words) from the document information D1 by a known morphological analysis method. The feature term extraction unit 132 calculates “specificity” for each of these terms. The specificity indicates the extent to which the term characterizes the content of the document information D1. The feature term extraction unit 132 sets a high degree of specificity for terms that frequently appear in the document information D1 and terms that are included in important parts such as titles and headings. The feature term extraction unit 132 may calculate the specificity of terms included in the document information D1 by applying a known document analysis method such as the TF/IDF (Term Frequency/Inverse Document Frequency) method. In the TF/IDF method, terms that are rarely used in general document groups (corpus) but appear in the document information D1 are determined to be highly specific terms.

The feature term extraction unit 132 identifies terms having a specificity equal to or higher than a predetermined value and ranked within the top five as feature terms indicating the characteristics of the document information. It is assumed that "Kyushu", "hot spring" and "digital art" are specified as characteristic terms from the document information D1.

Next, the coordinates (evaluation value) in the evaluation space of the characteristic term "Kyushu" are obtained. Here, let the coordinates of the important term “Korea” be (xa, ya) and the coordinates of the important term “shochu” be (xb, yb). The characteristic term evaluation unit 134 first calculates the degree of similarity between the characteristic term “Kyushu” and the important term “Korea”. Based on a known algorithm known as "word2vec" in this embodiment, the feature term evaluation unit 134 calculates the similarity between the key term "Korea" and the feature term "Kyushu". The similarity calculation method is not limited to word2vec, and any method may be used as long as it is determined as an index value that indicates the meaning of words and the closeness of categories.

Assume that the similarity between the characteristic term "Kyushu" and the important term "Korea" is r1, and the similarity between the characteristic term "Kyushu" and the important term "konaka" is r2. In this case, the coordinates (xc, yc) of the characteristic term "Kyushu" are
xc=(r1*xa+r2*xb)/(r1+r2)
yc=(r1×ya+r2×yb)/(r1+r2)
calculated as In other words, the feature term "Kyushu" is located between the key terms "Korea" and "shochu" and is attracted to key terms with a high degree of similarity. In this way, the coordinates in the feature term evaluation space are determined based on the key term and its evaluation value (coordinates) and the similarity between the key term and the feature term.

The same is true when there are three or more important terms, and the coordinates of the characteristic term “Kyushu” are calculated based on the similarity to each of the plurality of important terms (“Korea”, “Shochu”, “Shimonoseki”, “Golf”, “British”, “Rock”). determined based on According to FIG. 9, the feature term "Kyushu" is similar to key terms such as "shochu", "shimonoseki" and "korea", but is not very similar to key terms such as "UK" and "rock". .
The feature term evaluation unit 134 also performs coordinate calculation for the feature terms “hot spring” and “digital art” in a similar manner.

The conformity determination unit 136 then calculates the coordinates of the document information D1 in the evaluation space (hereinafter referred to as "document coordinates"). In this embodiment, the conformity determination unit 136 sets the center of gravity of the coordinates of the three characteristic terms "Kyushu", "hot spring" and "digital art" to the document coordinates (x1, y1) of the document information D1. The document coordinates of the document information D1 may be determined based on the coordinates of the characteristic term, and need not be limited to the center of gravity of the coordinates of the characteristic term. For example, the document coordinates may be adjusted to approximate the coordinates of feature terms with relatively high specificity more than the coordinates of feature terms with relatively low specificity.

According to the above processing results, document information D1 satisfies x1 (favorite)>0 and y1 (interest)<0, so it is presumed to be "information that may be liked but is not likely to be of interest". . On the other hand, if the coordinates (x2, y2) of another document information D2 satisfy x2 (like)>0 and y2 (interest)>0, then the document information D2 is "favorite content and interested Information that is likely to be obtained" is conjectured. In this case, the document information D2 has a higher "suitability" than the document information D1.

When the document coordinate (xd, yd) of certain document information satisfies the condition that "xd is equal to or greater than the threshold value TX and yd is also equal to or greater than the threshold value TY", the matching determination unit 136 determines that the document information is compatible with the user's preference. judge. The thresholds TX and TY referred to here can be arbitrarily set, and the conformity determination unit 136 may change these thresholds based on the settings from the user.

FIG. 10 is a flow chart of the document analysis process and match determination process.
In FIG. 10, S20 to S26 correspond to the document analysis process, and S28 and S30 correspond to the match determination process. The document acquisition unit 142 regularly crawls the Internet to collect document information. The processing shown in FIG. 10 is executed each time document information is acquired.

(Document analysis process)
The feature term evaluation unit 134 extracts one or more feature terms from the newly acquired document information (S20). Next, the feature term evaluation unit 134 calculates similarities between all feature terms and all key terms (S22). The feature term evaluation unit 134 calculates the coordinate values of the feature terms in the evaluation space based on the coordinate values (evaluation values) of the key terms and the degree of similarity between the key terms and the feature terms (S24). In other words, the characteristic term evaluation unit 134 evaluates the characteristic term based on the evaluation value of the important term. The matching determination unit 136 calculates document coordinates based on the coordinate values of the feature terms (S26). In summary, the match determination unit 136 evaluates document information based on the evaluation of feature terms. The evaluation of feature terms is based on the evaluation of key terms. The selection of important terms and their evaluation reflect the user's preferences.

(Compliance judgment process)
Based on the document coordinates, the suitability determination unit 136 determines whether or not the document information matches the user's preferences (S28). When it matches (Y of S28), the match determining section 136 instructs the output section 120 to display this document information on the screen (S30). If not, the process of S30 is skipped. The document information may be stored in the data storage unit 116 of the user terminal 100 if it matches, and the document information may be discarded if it does not match.

The concierge function of the user terminal 100 (preference analysis device) has been described above based on the embodiment.
According to the present embodiment, the user can easily verbalize the "source of thoughts" stored in his or her mind through dialogue with the butler 152 . When the user inputs many terms by free association, the association setting unit 126 enriches the term association information indicating the connection of the terms. The important term identifying unit 128 finds out important terms that characterize the user's preferences from this term-related information. The butler 152 may present important terms to the user as "Your mind is overflowing with such words." At this time, the user can feel as if the butler 152 is helping him "discover himself."

Based on characteristic terms that characterize the document information and key terms that characterize the user's preferences, the conformity determination unit 136 determines whether or not the information that the user is consciously or unconsciously seeking is included in the document information. do. With such a concierge function, it is possible to rationally select document information that matches the user's taste.

As described with reference to FIG. 1(b), the concierge function is implemented in the user terminal 100, so there is no need to externally provide information that is deeply related to privacy, such as important terms. Therefore, the user can easily interact with the butler 152 with peace of mind. The user's sense of security with respect to the concierge function (butler 152) is also important for eliciting important terms close to the user's true intentions. The user can enjoy the process of the butler 152 gaining a deeper understanding of himself while casually interacting with the butler 152 in a playful way.

It should be noted that the present invention is not limited to the above-described embodiments and modifications, and can be embodied by modifying constituent elements without departing from the scope of the invention. Various inventions may be formed by appropriately combining a plurality of constituent elements disclosed in the above embodiments and modifications. Also, some constituent elements may be deleted from all the constituent elements shown in the above embodiments and modifications.

[Modification]
Various modifications other than those described above are conceivable.

In the present embodiment, it has been described that when a derived term is associated with a standard term, a relationship (link) is set between the standard term and the derived term. Relevance is not limited to such an association, and may be "a relationship in which a derived term is recollected based on a reference term". For example, terms having opposite, similar, subordinate concept, superordinate concept, exemplified, etc. relationships to the reference term may be derived terms.

The "terms" in this embodiment have been described assuming words such as "Kyushu" and "rock". The "terms" to be processed are not limited to words, and may be compound words or sentences.

In the present embodiment, the important term identification unit 128 has been described as identifying important terms based on the degree of relevance between a certain term and other terms. In addition, the important term identification unit 128 may identify a term that appears more than a predetermined number of times in the term-related information as an important term, and term 2 is derived from term 1, and term 1 is derived from term 2. Both terms 1 and 2 may be identified as important terms when, ie, when two-way association is established. In this way, the important term identifying unit 128 identifies important terms based on various modes of occurrence of terms in the term-related information, such as relationships with other terms, frequency of occurrence of terms, and the like, as well as the number of relationships (links). should be specified.

The butler 152 may elicit key terms by posing various questions to the user. For example, butler 152 prepares various questions such as "What is your favorite flower?", "What would you do if you had 10 billion yen?" Reference terms may be derived.

In this embodiment, it has been described that the user inputs terms by responding to dialogue with the butler 152 . In addition, the user's preferences may be searched for from text posted by the user on SNS (Social Networking Service). For example, when the user posts "I did a triathlon in Kyushu", the term acquiring unit 122 extracts "Kyushu" and "triathlon". The association setting unit 126 associates "Kyushu" and "triathlon" appearing in the same sentence and registers them in the term association information. Also, if there is a post saying "I had a drinking party after the triathlon", the "triathlon" and the "drinking party" are associated. In this way, the association setting unit 126 may update the term-related information based on the article posted by the user on the SNS. The association setting unit 126 may update the term association information based on both the dialogue with the butler 152 and the posted article.

In the present embodiment, the description has been given assuming an edge type in which the user terminal 100 is equipped with the concierge function. As a modification, a server type in which the server 102 is equipped with a concierge function can also analyze user preferences. In this case, the user may interact with the concierge function provided by the server 102 via the user terminal 100, and the concierge function of the server 102 may analyze important terms from the terms provided by the user.

In the present embodiment, it has been described that the user numerically inputs evaluation values for important terms. As a modification, the important term evaluation unit 130 may calculate the evaluation value of each important term based on a known decision-making model such as AHP (Analytic Hierarchy Process). Thus, the user may directly input the evaluation value, or the decision-making model may calculate the evaluation value based on the input from the user.

The important term evaluation unit 130 may calculate the evaluation value of the important term from the article posted on the SNS. For example, if there is a posted article "I want to go to Kyushu", the important term evaluation unit 130 may increase the evaluation value for the interest of the important term "Kyushu" based on the term "I want to go" indicating interest. good. As another example, if there is a posted article stating that "especially 80's rock has many great songs", the important term evaluation unit 130 may increase the evaluation values for "like" and "interest" for "rock". . Affirmative terms such as "amazing" and "masterpiece" may be set in advance to imply "like" and "interest". Also, the important term evaluation unit 130 may reduce the evaluation value of "interest" for "rock" from the posted article "I listened to nothing but 80's rock, so I think I should listen to something different these days." By performing natural language analysis on the posted article in this way, the important term evaluation unit 130 may appropriately increase or decrease the evaluation of the important terms included in the posted article.

In the present embodiment, two axes of favorable evaluation and interest evaluation are exemplified as evaluation axes of important terms or characteristic terms. Other evaluation criteria include "familiar/not familiar", "familiar/not detailed", "scary/not scary", and "fun/boring".

The feature term evaluation unit 134 may adjust the evaluation value of the important term based on the article posted to the SNS. For example, when the user has given the term “was” representing an experience to the important term “Okinawa”, such as “I went to Okinawa”, the feature term evaluation unit 134 determines that the important term “Okinawa” The evaluation value for "interest" may be decreased. According to such a control method, even when the user is interested in Okinawa, it is possible to quickly reflect in the evaluation space the psychological tendency of the user to lose interest when the user actually experiences Okinawa.

The user may input an instruction to stop or delete important terms. It is assumed that the evaluation input unit 124 has received from the user an instruction to “suspend for 6 months” with respect to the important term “lock”. In this case, the important term specifying unit 128 temporarily removes the term "lock" from being an important term. With such a setting, the user can temporarily suppress provision of document information related to "lock". For example, even if the user is interested in rock, if a large amount of textual information related to rock is provided, the user may feel sick to such textual information. According to such a control method, the user can inform the concierge function that "I like locks and am interested, but I would like you to refrain from providing information on locks for a while."

A user may temporarily change the evaluation value of an important term. It is assumed that the evaluation input unit 124 has received a setting from the user to the effect that "the evaluation value of 'interest' will be zero for 6 months" for the important term "UK". In this case, the important term evaluation unit 130 sets the interest-related evaluation value of the important term “UK” to zero for only six months, and restores the original evaluation value after six months. Such a control method also enables the user to inform the concierge function of temporary changes in interest.

The association setting unit 126 may update the term association information over time. For example, of terms 1, 2, and 3 acquired in the first period, term 1 may be deleted from the term-related information when term 1 is not reacquired in the subsequent second period. At this time, the important term specifying unit 128 re-specifies the important term based on the updated term-related information. According to such a processing method, the user can effectively exclude terms that no longer occupy his mind from the term-related information while repeatedly interacting with the butler 152 .

100 user terminal, 102 server, 104 external database, 106 internal database, 110 user interface processing unit, 112 data processing unit, 114 communication unit, 116 data storage unit, 118 input unit, 120 output unit, 122 term acquisition unit, 124 evaluation input unit, 126 relation setting unit, 128 important term identification unit, 130 important term evaluation unit, 132 feature term extraction unit, 134 feature term evaluation unit, 136 match determination unit, 138 transmission unit, 140 reception unit, 142 document acquisition unit, 150 dialogue screen, 152 butler, 154 term input area, 156 reply button

Claims (5)

a term acquisition unit that receives an input of a derivative term recollected from the reference term from the user to be analyzed;
a relevance setting unit that updates term-related information indicating relevance of terms by associating a reference term with a derived term;
an important term identification unit that identifies important terms for the user,
The association setting unit updates the term-related information when a new derivative term is acquired after setting the derivative term as the next reference term,
The preference analysis device, wherein the important term identification unit identifies the important term from the plurality of terms based on the manner of appearance of each of the plurality of terms included in the term related information. The important term identification unit determines whether or not the first term is an important term based on the number of relationships of the first term to be verified with other terms. The preference analyzer according to claim 1. 3. The preference analysis device according to claim 1, further comprising an important term evaluation unit that gives evaluation values to important terms. a document acquisition unit that acquires document information;
a feature term extraction unit that extracts feature terms from the document information;
a feature term evaluation unit that calculates an evaluation value of the feature term based on the similarity between the key term for the user and the extracted feature term and an evaluation value of the key term;
4. The preference analysis apparatus according to claim 3, further comprising a compatibility determination unit that determines compatibility between the document information and the user based on evaluation values of the characteristic terms. A function to receive input of derived terms recalled from the reference term from the user to be analyzed;
the ability to update term-related information that indicates the relevance of terms by associating base terms and derived terms;
A function of updating the term-related information when a new derived term is acquired with the derived term as the next reference term;
A preference analysis program for causing a computer to exhibit a function of identifying important terms for the user from the plurality of terms based on the appearance of each of the plurality of terms included in the term-related information.

Images