GB2517358A

GB2517358A - Recommendation creation system

Info

Publication number: GB2517358A
Application number: GB1421558.6A
Authority: GB
Inventors: Tomoya Ishida; Akihiro Nishiyama; Naoharu Takata; Yosuke Maki; Yukihiro Horikita; Rintaro Washio
Original assignee: Dentsu Inc
Current assignee: Dentsu Group Inc
Priority date: 2013-02-01
Filing date: 2013-02-01
Publication date: 2015-02-18
Also published as: JP5782562B2; WO2014118975A1; GB201421558D0; JPWO2014118975A1

Abstract

The objective of the present invention is to provide a recommendation creation system capable of providing responses not expected by the user or optimal recommendations such as responses appropriate for the user. This recommendation creation system is characterized in being equipped with: a database which stores user information sorted according to a plurality of categories which each include at least one item; a question control means for generating a question to the user for acquiring user information related to uncollected items in the categories; and a recommendation creation means for creating recommendations for queries by the user on the basis of the user information related to items in the plurality of categories stored in the database.

Description

DESCRI PTION

RECOMMENDATION CREATING SYSTEM

Technical Field

[0001] The present invention relates to a recommendation creating system that provides an optimum answer to a user's question.

Background Art

[0002] In recent years, services are provided which analyze a user's speech inputted from a mobile phone or the like, and search and acquire information requested by the user. These services provide information or the like in accordance with user characteristics for the purpose of providing an accurate answer or optimum answer to the user. As a technique for implementing such services, for example, Patent Literature 1 discloses a portable information terminal provided with an agent function that enables search processing on information necessary for contents of a user's action request.

Citation List Patent Literature [0003] Patent Literature 1 Japanese Patent Application Laid-Open No.10-283362

Summary of Invention

Technical Problem [0004] However, since the above-described information search services provide answers from keywords and the user's attribute information included in the user's query, answers unexpected by the user or recommendations appropriate to the user cannot be obtained.

[0005] The present invention has been implemented in view of the above problems, and it is an object of the present invention to provide a recommendation creating system capable of providing optimum recommendations such as answers unexpected by the user or answers appropriate to the user.

Solution to Problem [0006] A recommendation creating system of the present invention includes a database that stores a user's information classified under a plurality of categories, each of which includes at least one item, question control means for generating a question to the user to acquire the user information relating to uncollected items in the categories, and recommendation creating means for creating a recommendation in response to the user's query from the user information relating to the items in the plurality of categories stored in the database.

[0007] According to this configuration, it is possible to provide optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0008] In the recommendation creating system of the present invention, the question control means preferably estimates one category from a keyword included in the query, assumes the estimated category as a central category, assumes categories other than the central category related to the query as related categories and sets a block matrix using the central category and the related categories.

[0009] According to this configuration, it is possible to use user information for items about categories other than a category estimated from the user's query, and thereby provide optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0010] In the recommendation creating system of the present invention, a degree of relatedness to other categories is set beforehand in respective items included in each category, and the question control means preferably generates a question to acquire user information relating to uncollected items in the related categories having a high degree of relatedness to the central category.

[00111 According to this configuration, it is possible to collect uncollected items with a high degree of relatedness to the central category first, and thereby provide more optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0012] In the recommendation creating system of the present invention, when a registration rate of items in the central category and related categories making up the block matrix is less than a predetermined value, the question control means preferably generates a question about items for which user information is uncollected.

[0013] According to this configuration, it is possible to provide recommendations including information on related categories in response to the user's query, and thereby obtain answers which are so unexpected that the user himself/herself is S even unaware of and which are beneficial to the user.

[0014] In the recommendation creating system of the present invention, the recommendation creating means preferably creates recommendations in response to the user's query also using the user's history log.

[0015] According to this configuration, since the user's history log is used as the user information, it is possible to obtain answers which are so unexpected that the user himself/herself is even unaware of and which are beneficial to the user, and provide more optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0016] In the recommendation creating system of the present invention, the database preferably stores questions to acquire user information relating to each item in each category.

[0017] In the recommendation creating system of the present invention, the database preferably stores a block matrix set for the each central category.

[0018] A portable terminal apparatus of the present invention has a feature of mounting the above-described recommendation creating system.

Advantageous Effects of Invention [0019] When providing a recommendation in response to a query from a user, the present invention uses items with a high degree of relatedness to the query first, asks the user a question, when there are uncollected items despite a high degree s of relatedness, to collect such items, and can thereby provide optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

Brief Description of Drawings

[0020] Figure 1 is a diagram illustrating a system configuration including a recommendation creating system according to an embodiment of the present invention; Figure 2 is a block diagram illustrating a schematic configuration of the recommendation creating system according to the embodiment of the present invention; Figure 3 is a diagram illustrating an example of a block matrix; Figure 4 is a diagram illustrating an example of a degree of relatedness to a central item in the block matrix; Figure 5 is a diagram illustrating an example of a block matrix including a history log; Figure 6 is a diagram illustrating examples of recommendation pattern; Figure 7 is a diagram illustrating time transition of an amount of deposit; Figure 8 is a flowchart illustrating a recommendation creating operation in 23 the recommendation creating system according to the embodiment of the present invention; Figure 9 is a diagram illustrating another example of the block matrix; and Figure 10 is a diagram illustrating examples of recommendation pattern.

Description of Embodiments

[0021] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Figure 1 is a diagram illustrating a system configuration including a recommendation creating system according to an embodiment of the present invention. Figure 1 shows a case where the recommendation creating system is located on a server connected to a network. The present invention is not limited to the system configuration shown in Figure 1, but is also applicable to a case where the recommendation creating system is mounted on a portable terminal apparatus.

[0022] The system shown in Figure 1 is mainly constructed of a portable terminal apparatus 1 such as a mobile phone, a search engine 4 connected to this portable terminal apparatus 1 via a mobile network 2 and a network 3 such as the Internet, and a recommendation creating system (Web server) 5 connected via the mobile network 2 and the network 3 such as the Internet.

[0023] The portable terminal apparatus 1 includes a terminal apparatus such as a mobile phone or tablet type terminal, and is provided with functions at least enabling it to access the search engine 4 to perform information search and enabling it to access the server provided with the recommendation creating system to acquire recommendations. The portable terminal apparatus 1 is also provided with a function of analyzing a user's audio input speech and acquiring information requested by the user from the search engine 4. Moreover, the portable terminal apparatus 1 may also be provided with other functions such as a speech function, electronic mail function and non-contact card function.

[0024] The mobile network 2 is a network between the portable terminal apparatus 1 and the Internet 3. In addition to a normal mobile communication network, this mobile network 2 also includes a mobile packet communication network. The mobile network 2 carries a gateway function of providing connections with the network 3 such as the Internet.

[0025] The network 3 includes not only the Internet but also other networks such as LAN and WAN.

[0026] The recommendation creating system 5 is mainly constructed of an interaction control section 51, a recommendation creating section 52, a question control section 53, and a personal information database (DB) 54 as shown in Figure 2. When providing a recommendation in response to a query from the user, the recommendation creating system 5 provides a recommendation using items with a high degree of relatedness to the query first. Furthermore, if there are uncollected items despite having a high degree of relatedness, the recommendation creating system 5 asks the user a question about collecting such items.

[0027] The interaction control section 51 not only outputs to the user a recommendation in response to the query inputted from the user but also outputs to the user a question for acquiring user information relating to uncollected items.

Here, the "query" refers to a question inputted by the user to the recommendation creating system to ask for a recommendation and the "recommendation" refers to a proposal provided by the recommendation creating system in response to the "query' from the user. When, for example, the user inputs to the recommendation creating system as the "query," "I'm considering purchasing a house. What do you think.,""the debt is big compared to the amount of deposit. What if you first sell your car for which you have to pay the remaining loan?" is outputted as a "recommendation." [0028] Furthermore, the "question" refers to a question which is stored in the personal information DB 54 which will be described later and which corresponds to an uncollected item of user information classified by category including at least one item. For example, in a category of financial information, if an "amount of deposit" item is uncollected, a question "What is the amount of deposit?" is outputted from the recommendation creating system to the user.

[0029] The "category" is intended to classify user information, and includes, for example, food, health, human relationship, dream/goal, basic information, hobby, sports, job and financial. The "item" refers to specific information included in a category. An example of item included in the food category is favorite food and an example of item included in the financial category is an amount of deposit, regular expenses or amount of debt. The basic information is basic information on a user such as name, height, weight, address, age, occupation and family composition, and this basic information may be inputted by the user beforehand or information inputted by the user may be used during operation of another application or may be collected by questions. These categories and items are not limited to those described above, but can be set as appropriate.

[0030] The interaction control section 51 estimates a category under which user information is classified from keywords included in a query inputted from the user.

For example, when the query from the user is "I'm considering purchasing a house.

What do you think?," a "financial category" is estimated as a central category from "house" and "purchasing" which can be keywords. The interaction control section 51 outputs the estimated central category to the question control section 53.

When the user inputs data to the interaction control section 51, the user may input the data by means of speech or input the data manually.

[00311 The interaction control section 51 changes outputs of the above-described questions and outputs of the above-described recommendations in accordance with a collection situation of user information (registration rate) regarding items in a block matrix set with the above-described central category and related categories other than this central category. That is, when the registration rate in the block matrix is a predetermined value (e.g., 80%) or above, the interaction control section 51 outputs a recommendation in response to the query from the recommendation creating section 52 which will be described later. Note that whether or rot the registration rate is a predetermined value or above is notified from the question control section 53 which will be described later. On the other hand, when the registration rate in the block matrix is less than the predetermined value (e.g., 80%), the interaction control section 51 outputs questions corresponding to uncollected items from the question control section 53.

[0032] The question control section 53 generates a question to the user to acquire user information relating to uncollected items in the central category and related categories (block matrix). Furthermore, the question control section 53 estimates one category from keywords included in the query, assumes the estimated category as a central category, assumes categories other than the central category related to the query as related categories and sets the block matrix with the central category and related categories. By using the block matrix in this way, it is possible to use user information corresponding to items about categories other than the category estimated from the user's query, and thereby provide optimum recommendations such as answers unexpected by the user and answers appropriate to the user or the like.

[0033] For example, since a central category is notified from the interaction control section 51 as described above, the question control section 53 sets a block matrix as shown in Figure 3 composed of this central category A as the center and related categories B which are categories other than the central category A related to the question arranged around the central category A. In Figure 3, the central category A is a financial category and the related categories B are food category, hobby category, sports category and health category. Note that this block matrix is set beforehand for each type of central category and the block matrix is designed to be determined when the central category is determined. The block matrix set for each central category is stored in the personal information DB 54.

[0034] Furthermore, the question control section 53 generates a question to acquire user information relating to uncollected items having a high degree of relatedness to the central category. Note that the degree of relatedness is set for each block matrix. For each item of the central category and related categories, a degree of relatedness (distance) from the central item of the central category is set beforehand, and therefore the question control section 53 generates questions about items with a high degree of relatedness (short distance) first. For example, as shown in Figure 4, the distance from a central item of the central category A (central item in the central category (distance 0)) to another item is set in the block matrix. The question control section 53 generates questions about items in the selected block matrix for which user information is uncollected and which have small distances from the central item of the central category A (high degree of relatedness to the central item) first. This makes it possible to collect uncollected S items with a high degree of relatedness to the central category with higher priority than uncollected items with a low degree of relatedness to the central category, and thereby provide more optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0035] Questions to be generated are stored beforehand in the personal information DB 54 by category or by item as questions to acquire user information relating to each item in each category, and it is possible to read the questions by specifying an item and category. For example, questions are managed using a data record [item, category, question] and can be stored in the personal information is DB 54 after sorting the questions into a list (table) using the item as a first key and the category as a second key in that order. Flags are added to this list for each item to indicate whether the user information is already collected or uncollected and a user information collection situation about an item is also associated therewith. For example, if the user information is already collected, flag 1" is inputted and if the user information is uncollected, "0" is inputted.

[0036] Although a case has been described in the present embodiment where questions are stored beforehand in the personal information DB 54, the present invention is not limited to this, and the present invention is also applicable to a case where the question control section generates questions about uncollected items of the block matrix stored in the personal information DB 54.

[0037] The question control section 53 searches this list through a binary search or linear search using an item determined by a distance from the central item as a first key and the central category as a second key as described above, and generates a question with a matching record as a question to be outputted to the user. Contents of the question are not particularly limited as long as such contents allow user information on uncollected items to be collected.

[0038] When the registration rate of items in the central category and related categories making up the block matrix is less than a predetermined value, the question control section 53 generates questions about items for which user information is uncollected. Here, the "registration rate" refers to a ratio of the number of items for which user information is already collected (number of items with flag "1") to the total number of items in the central category and related categories making up the block matrix (45 items in Figure 3). For example, when the registration rate is less than 80%, if a setting is made such that questions are generated about items for which user information is uncollected, the question control section 53 continues to generate questions about items for which user information is uncollected until the registration rate becomes 80% or more. In Figure 5, items for which user information is already collected (e.g., "amount of deposit" item, "amount of debt" item, "regular expenses" item of the financial category) are shaded. Therefore, questions are sequentially asked about items for which user information is uncollected (items not shaded in Figure 5) until the registration rate of items becomes 80% or more.

[0039] Although the present embodiment has described a case where a recommendation is created when the registration rate of the user information becomes a predetermined value or above, the present invention is not limited to this, and may be configured to assign weights to items in a category and create a recommendation when user information on all heavily weighted items (of high importance) is collected. Even when the registration rate is less than 80%, a setting may be made so that if the registration rate is within a certain range (e.g., 60% to 80%), a question about uncollected items is outputted to the user and a recommendation is outputted as well.

[0040] When a question is generated, an answer is received from the user and user information on the items is collected, the question control section 53 calculates the registration rate, and outputs, when the registration rate becomes 80% or more, such information to the interaction control section 51 and the recommendation creating section 52. This means that no recommendation is outputted to the user until the registration rate becomes 80% or more. The threshold of the registration rate is not limited to this, but can be set as appropriate.

[0041] Thus, when the registration rate becomes a predetermined value or above, a recommendation is outputted, and it is thereby possible to provide a recommendation including information on related categories in response to the user's query and thereby obtain answers which are so unexpected that the user himself/herself is even unaware of and which are beneficial to the user.

[0042] The personal information DB 54 stores the user's information classified into a plurality of categories, each of which includes at least one item. The personal information DB 54 also stores a block matrix set for each central category. The personal information DB 54 also stores questions for acquiring user information relating to each item in each category.

[00431 Moreover, the personal information DB 54 stores a recommendation pattern in accordance with a combination of items for which user information is already collected. This recommendation pattern is predetermined for each set block matrix. As shown in a table in Figure 6, the recommendation pattern is associated with items for which user information is already collected. For example, recommendation pattern I is a pattern of creating a recommendation from items such as "amount of deposit" item, "history of amount of deposit" item, "regular expenses" item (financial category), "car item" (hobby category), "favorite food" item, "food history item" (food category). Recommendation pattern 2 is a pattern of creating a recommendation from "amount of deposit" item, "car" item (hobby category), "favorite food" item, and other search information. These recommendation patterns are not limited to Figure 6 and can be set as appropriate.

[0044] The recommendation creating section 52 creates a recommendation in response to a query from the user from the user information relating to items in the plurality of categories stored in the personal information DB 54. Upon receiving such notification that the registration rate has become a predetermined value or above from the question control section 53, the recommendation creating section 52 creates a recommendation in response to the user's query. In this case, the recommendation creating section 52 creates a recommendation for the user using the user information about items for which user information is already collected in the block matrix. When creating a recommendation, the recommendation creating section 52 may also create a recommendation using only items for which user information is already collected in the block matrix or create a recommendation using other search information (e.g., information related to items for which user information is already collected) in addition to the items for which user information is already collected in the block matrix.

[0045] For example, the recommendation creating section 52 selects a recommendation pattern from items for which user information is already collected and creates a recommendation using the recommendation pattern. In this case, the recommendation creating section 52 collects information using the search engine 4 for information other than the collected user information and necessary for the recommendation. The recommendation creating section 52 creates a recommendation using such user information and search information.

[0046] As shown in Figure 5 and Figure 6, the recommendation creating section 52 creates a recommendation for the user's query using the user's history log as well. As shown, for example, in Figure 7, this user's history log shows a time course regarding an item in a category (Figure 7 shows a time course of the amount of deposit). Using the user's history log as user information in this way, it is possible to obtain answers which are so unexpected that the user himself/herself is even unaware of and which are beneficial to the user and provide more optimum recommendations such as answers unexpected by the user and answers appropriate to the user.

[0047] A recommendation creating procedure by the recommendation creating system in the above-described configuration will be described using Figure 8.

[0048] First, the interaction control section 51 estimates a category under which user information is classified from keywords included in a query inputted by the user (5Th). For example, when the query from the user is I'm considering purchasing a house. What do you think?," the interaction control section 51 estimates a "financial category" as a central category from "house" and "purchasing" which can be keywords. Information of the estimated category is outputted to the question control section 53.

[0049] Next, the question control section 53 sets a block matrix whose central category is a financial category (ST12). This block matrix is set by selecting a preset block matrix stored in the personal information DB 54. Next, a registration rate of the set block matrix is obtained. The registration rate is obtained by calculating a ratio of the number of items for which user information is already collected to the total number of items in the central category and related categories making up the block matrix. The question control section 53 determines whether or not the registration rate is equal to or above a predetermined value (here 80%) (STI 3).

[0050] When the registration rate is less than 80% (N), the question control section 53 generates a question to the user to acquire user information relating to uncollected items in the block matrix. First, an item is selected which is an item for which user information is uncollected in the block matrix and which has a small distance from a central item of a central category (high degree of relatedness to the central item) (ST14). The question control section 53 generates a question about the selected uncollected item. To be more specific, the question is extracted using the item and category as keys in the list managed using data records [item, category, question]. The extracted question is outputted to the user via the interaction control section 51 (STI5).

[0051] Next, when the user answers the question, the answer is sent to the personal information DB 54 via the interaction control section 51 and registered with an item of the block matrix (selected uncollected item) as user information (STI6). In this case, a flag of the item for which user information is registered is rewritten (uncollected -* collected) in the personal information DB 54.

[0052] Next, the question control section 53 obtains the registration rate of the block matrix after registering the user information and determines whether or not the registration rate is equal to or above a predetermined value (here 80%) (S113).

When the registration rate is less than 80% (N), the procedure in ST14 to S116 is repeated. On the other hand, when the registration rate is 80% or more (Y), such information is notified from the question control section 53 to the recommendation creating section 52, and the recommendation creating section 52 creates a recommendation on the user's query (S117). In this case, the recommendation creating section 52 selects a recommendation pattern from among the items for which user information is already collected with reference to the table stored in the personal information DB 54. The recommendation creating section 52 then creates a recommendation according to this recommendation pattern. The created recommendation is outputted to the user via the interaction control section 51.

[0053] The present embodiment has described a case where questions to the user are repeated until the registration rate becomes a predetermined value or above, but the present invention is not limited to this, and the interaction control section 51 may output a search result other than questions to the user. For example, after the user makes a "query" to the system, the system may output the question to the user, collect user information of the uncollected item and output the search result in response to a normal search request from the user using a search engine. In this case, the interaction control section 51 may combine the question to the user and the search result, and output the combination to the user [0054] Hereinafter, examples implemented to define effects of the present invention will be described. However, the present invention will be by no means limited to the following examples.

[0055] <Example of user considering purchasing house> When the user inputs to the portable terminal apparatus "I'm considering purchasing a house. What do you think?," this query is sent to the recommendation creating system via the network. The recommendation creating system estimates a category from the user's query and determines a block matrix.

Here, the recommendation creating system estimates the category to be a financial category and determines a block matrix with the financial category as the central category (Figure 3).

[0056] is The recommendation creating system asks the user a question about an item with a high degree of relatedness to the central item of the central category while monitoring the registration rate and collects user information of the uncollected item. This operation is continued until the registration rate becomes a predetermined value or above. Then, when the registration rate becomes a predetermined value or above, the recommendation creating system creates a recommendation using the collected user information and search information.

[0057] In this case, the recommendation creating system selects a recommendation pattern from among the collected items and creates a recommendation according to the recommendation pattern. The recommendation pattern uses at least the "amount of deposit" item, history of amount of deposit" item, "regular expenses" item (financial category), "car" item (hobby category), favorite food" item, and "food history" item (food category) (recommendation pattern I in Figure 6). Here, the history of amount of deposit is assumed to be as shown in Figure 7. Furthermore, the recommendation creating system searches the price of the house or the like using the search engine 4.

[0058] The recommendation created in this way is as follows and* this recommendation is outputted to the user The price of the house is aboutXyen, but since the amount of deposit isY yen, it is not enough. However, you should be careful because your expenses are growing. By the way, since you have a car, you generally spendZyen a month and because your favorite food is 0 which costs a lot, you should not buy the house now!." [0059] According to the recommendation creating system of the present invention, instead of making a recommendation introducing a house in response to the query of considering purchasing a house, an unexpected answer is obtained which recommends abstaining from purchasing a house.

[0060] <Example of user who wants to improve high blood pressure> When the user inputs to the portable terminal apparatus "I want to reduce my blood pressure down to 145 mmHg or less, what should I do?.," this query is sent to the recommendation creating system via a network. The recommendation creating system estimates a category from the user's query and determines a block matrix. Here, the recommendation creating system estimates the category to be a health category and determines a block matrix whose central category is a health category (Figure 9).

[0061] The recommendation creating system then asks the user a question about an item having a high degree of relatedness to the central item of the central category while monitoring the registration rate and collects user information of the uncollected items. This operation is continued until the registration rate becomes a predetermined value or above. After that, when the registration rate becomes a predetermined value or above, the recommendation creating system creates a recommendation using collected user information and search information.

[0062] In this case, the recommendation creating system selects a recommendation pattern from the collected items and creates a recommendation according to the recommendation pattern. This recommendation pattern uses at least a blood pressure item, cigarette item (health category), TOEIC item (job category), jogging item (sports category), and alcohol item (food category) (recommendation pattern 3 in Figure 10).

is [0063] The recommendation created in this way is as follows and this recommendation is outputted to the user.

"Though you puff a cigarette, you don't do jogging or other exercise, and moreover you seem to like alcohol -that's a problem. By the way, you seem to take a considerable amount of alcohol on your day off, why not study TOEIC spending that time?." [0064] Thus, according to the recommendation creating system of the present invention, an unexpected answer of studying TOEIC is obtained for the query "I want to reduce my blood pressure." [0065] The present invention is not limited to the above-described embodiments, but may be implemented modified in various ways. The block matrix, category, item, query, question or the like can be implemented modified in various ways as appropriate without departing from the scope of the present invention. Other parts may also be implemented modified in various ways as appropriate without departing from the scope of the present invention.