JP6752330B1 - Machine learning-based matching equipment and matching methods - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recommend an expert who is most suitable for a consultant. A matching device for matching a user with an expert extracts demographic information of the user and a first weighted key phrase based on the received user information. The device reads a second weighted keyphrase associated with demographic information in the same category as the user's demographic information from the trained data and compares the first weighted keyphrase with the second weighted keyphrase. From the experts associated with the topic estimated in the above, the expert selected according to the predetermined criteria is recommended to the user. [Selection diagram] Fig. 6

Description

The present invention relates to a machine learning based matching device and matching method. More specifically, the present invention relates to a machine learning-based matching device and matching method that recommends the optimum FP for the consultant based on the consultant information and the FP information. The present invention also relates to a machine learning-based matching device and matching method that extracts a new key phrase based on information generated during a chat between a consultant and an FP and provides a changed topic. ..

When discussing money and life plans, the number of people who consult with a specialist financial planner (FP) is gradually increasing. The counselor can consult at free counseling meetings held in various places, or the site (non-patent) of the Japan Association for Financial Planners (NPO) Japan Financial Planners Association (referred to as "Japan Association for Financial Planners" in this specification). In Reference 1), enter the desired conditions, search for an FP that meets the conditions, and make contact with them for consultation.

Consultation with the FP may be conducted face-to-face after contacting in advance, or may be conducted via telephone or electronic media (for example, email, chat, etc.).

When a counselor wants to talk to an FP, he or she may know that he or she wants to talk, but may not be aware of the potential needs. Also, from the standpoint of FP, the most suitable consultation content (for example, age, gender, residential area, educational background, family structure, etc.) depends on the attributes of the counselor, rather than the topics that the counselor actually wants to consult. , Asset formation such as savings is more important than buying a house, etc.), and the content of the consultation may be different.

Japan Association for Financial Planners, Finding a Financial Planner to Consult, [online], [Search May 15, 2019], Internet <URL: https://www.jafp.or.jp/confer/search/cfp/>

In the prior art, a scheme has been provided in which a counselor actively selects an FP as a counselor based on limited information. For this reason, there is a problem that an FP suitable for the potential needs of the counselor cannot be selected, and / or an FP who is good at counseling contents more appropriate for the counselor's attributes cannot be consulted. In addition, in the conventional scheme, it was not possible for the FP side to select a counselor who fits his or her specialty.

Conventionally, the counselor often consults with the FP first by e-mail or chat, and then actually consults face-to-face if the feeling is good, rather than consulting face-to-face from the beginning. Since each FP also has its own field of expertise, it may not be possible for the counselor to provide the optimal counseling simply by actively selecting the FP, which reduces the counselor's satisfaction. There were many cases where face-to-face consultation could not be actually conducted.

Furthermore, it is possible that an important topic for the counselor may be derived from the content of the conversation during the counseling, but some counselors and FPs may not notice the topic.

The present invention has been made to solve such a problem, and is a machine learning-based matching device and a matching method that recommends the optimum FP for the counselor simply by inputting predetermined information. The purpose is to provide. The present invention also provides a machine learning-based matching device and matching method that extracts a new key phrase based on information generated during a chat between a consultant and an FP and provides a changed topic. The purpose is to do.

A matching device for matching a user and an expert, which is one aspect of the present invention, receives user information from a first terminal device associated with the user and categorizes the received user information into a predetermined format. By doing so, the demographic information of the user is generated, and the first weighted key phrase is extracted by using a predetermined algorithm based on the information of the consultation content included in the received user information. , The second weighted keyphrase associated with the demographic information of the same category as the demographic information of the user is read from the trained data, and the first weighted keyphrase and the second weighted keyphrase are obtained. Estimate the topic by comparison, obtain the list of the experts associated with the estimated topic, and select the expert from the obtained list of the experts according to a predetermined criterion. It is characterized in that it is configured to recommend the selected expert to the user.

Another aspect of the present invention, the matching method performed by the matching device for user-expert matching, is to receive and receive user information from a first terminal device associated with the user. By categorizing the user information in a predetermined format, the demographic information of the user is generated, and a predetermined algorithm is used based on the information of the consultation content included in the received user information. Then, the first weighted key phrase is extracted, and the second weighted key phrase associated with the demographic information in the same category as the demographic information of the user is read from the learned data, and the first weight is obtained. Estimating a topic by comparing the attached keyphrase with the second weighted keyphrase, obtaining a list of the experts associated with the estimated topic, and obtaining the expert's It is characterized in that it comprises selecting an expert from a list according to a predetermined criterion and recommending the selected expert to the user.

According to the present invention, it is possible to recommend the optimum FP for the counselor, and to extract an important topic from the conversation between the counselor and the FP even during the consultation and provide a changed topic. It will be possible. This makes it possible to extract the potential needs of the counselor, thereby providing the counselor with more appropriate advice from the FP.

A detailed understanding of the embodiments disclosed herein can be obtained from the following description illustrated in connection with the accompanying drawings.
It is a block diagram of the whole system including the apparatus which concerns on embodiment of this invention. It is a system block diagram of the apparatus which concerns on embodiment of this invention. It is a figure which shows an example of the data structure of the consultant information which concerns on embodiment of this invention. It is a figure which shows an example of the data structure of the FP information which concerns on embodiment of this invention. It is a figure which shows the image of the structured data. It is a figure explaining the FP recommendation processing and the machine learning processing executed by the apparatus which concerns on embodiment of this invention. It is a figure explaining the topic change process executed by the apparatus which concerns on embodiment of this invention. It is a figure which shows an example of the screen of a chat room.

In the present specification, an embodiment in which the consultant and the FP consult by chat will be described as an example, but the technical idea which is the gist of the present invention can be applied to other technical means. That is, as an example of an expert, FP is mentioned in this specification, but a scheme that recommends other experts (for example, a legal expert, a tax / accounting expert, and an expert in various fields). The present invention can also be applied to. Hereinafter, embodiments of the present invention will be described in detail.

(overall structure)
FIG. 1 is a configuration diagram of the entire system including the device 10 according to the embodiment of the present invention. The device 10 is communicably connected to one or more consultant terminals 11, one or more FP terminals 12, and one or more external systems 13 via a network. In the present specification, the device 10 is described as one device, but various processes executed by the device 10 may be configured to be distributed and executed by a plurality of devices. In FIG. 1, only one consultant terminal 11, FP terminal 12, and external system 13 are shown, but there may be a plurality of these.

The device 10 is a machine learning-based matching device that receives predetermined information from the counselor terminal 11 and the FP terminal 12 and recommends the optimum FP for the counselor based on the received information. The device 10 is a machine learning-based matching device that extracts a new key phrase based on the information of the chat between the counselor and the FP and provides the changed topic.

The counselor terminal 11 is a terminal used by a user who wants to consult with the FP. The FP terminal 12 is a terminal used by the FP to be consulted. The counselor terminal 11 and the FP terminal 12 are computers that can access the device 10, and can use a real-time messaging tool such as chat. The counselor terminal 11 and the FP terminal 12 can be, for example, a computer having a communication function such as a personal computer (PC), a smartphone, or a tablet terminal, but are not limited to a specific device.

The external system 13 is a system that stores consultation and response information on various topics for each demographic information and provides the information to the outside. The external system 13 can provide the stored information to the device 10 in response to a request from the device 10. Demographic information refers to demographic attributes, that is, data on a person's socio-economic characteristics such as gender, age, area of residence, income, occupation, academic background, and family structure. Data analysis can be performed based on demographic information from the idea that the behavior / attitude of a person with that attribute differs depending on the demographic information.

The device 10 can receive feedback of the consultation result with the FP and the interview information when the face-to-face consultation with the FP is actually performed from the counselor terminal 11.

(System configuration)
FIG. 2 is a system configuration diagram of the device 10 according to the embodiment of the present invention. As shown in FIG. 2, the device 10 has a control unit 101, a main storage unit 102, an auxiliary storage unit 103, an interface (IF) unit 104, and an output, which are connected to each other by a bus 120 or the like, like a general computer. A unit 105 is provided. The device 10 also includes consultant information 106, FP information 107, and structured learning data 108 in the form of a file / database or the like.

The control unit 101, also called a central processing unit (CPU), controls each component of the device 10 and calculates data, and reads various programs stored in the auxiliary storage unit 103 into the main storage unit 102. And execute. The main storage unit 102, also called a main memory, stores various received data, computer-executable instructions, and data after arithmetic processing by the instructions. The auxiliary storage unit 103 is a storage device typified by a hard disk (HDD) or the like, and is used when storing data or programs for a long period of time.

The embodiment of FIG. 2 describes an embodiment in which the control unit 101, the main storage unit 102, and the auxiliary storage unit 103 are provided inside the same computer, but as another embodiment, the device 10 has the control unit 101, the main storage unit 101. By using a plurality of storage units 102 and auxiliary storage units 103, it is possible to configure parallel distributed processing by a plurality of computers. Further, as another embodiment, it is also possible to install a plurality of servers for the device 10 so that the plurality of servers share one auxiliary storage unit 103.

The IF unit 104 serves as an interface for transmitting and receiving data to and from other systems and devices, and also provides an interface for receiving various commands and input data (various masters, tables, etc.) from the system operator. The output unit 105 provides a display screen for displaying the processed data, a printing means for printing the data, and the like.

The counselor information 106 is a file that stores information received from the counselor, demographic information of the counselor, key phrases related to the consultation content, ranked topics derived from each key phrase, and feedback information of the consultation result. .. FIG. 3 is a diagram showing an example of the data structure of the consultant information 106 according to the embodiment of the present invention. The counselor information 106 can include the counselor ID 301, the counselor reception information 302, the counselor demographic information 303, the key phrase 304, the topic 305, and the feedback information 306, but is limited to these data items. It is possible to include other data items (eg, consultation date, etc.).

The counselor ID 301 is an identifier that identifies the counselor. The counselor reception information 302 indicates information received from the counselor, for example, information such as gender, age, presence / absence of spouse / children, residential area, occupation, annual income, financial assets held, and consultation content. The counselor demographic information 303 indicates the demographic information of the counselor generated based on the counselor reception information 302.

The key phrase 304 indicates a key phrase related to the consultation content extracted from the consultant reception information 302 by the algorithm of PositionRank and EmbeddedRank. For example, the key phrase 304 may refer to terms such as mortgage, floating interest rate, savings, health insurance, and medical expenses.

PositionRank is an algorithm for extracting a key phrase (a phrase representing a topic of a document) from a document. In PositionRank, a graph showing co-occurrence information of words is generated by morphological analysis of a document, the importance of each word is calculated from the appearance position of the word, and phrases with high importance are extracted. EmbedRank is an algorithm that extracts important key phrases contained in a sentence without preparing teacher data. In EmbeddedRank, candidate phrases are extracted from the sentence based on the information of part of speech (for example, adjective + noun), and the sentence and the vector of each phrase are acquired by using sentence embedding (Sentence Embedding). The similarity to the embedded vector is used to rank candidate phrases and determine weighted key phrases.

Topic 305 shows the ranked key phrases after matching each key phrase of the key phrase 304 with the information stored in the structured learning data 108. That is, in topic 305, the key phrases of the consultation contents of the counselor are weighted based on the information of the structured learning data 108, and it is estimated that the topic 305 is highly important from the demographic information to which the counselor belongs. Indicates a key phrase. Therefore, the information on topic 305 may be narrower than the information on key phrase 304, or may be the same as the information on key phrase 304. The counselor demographic information 303 and the key phrase 304 information of the counselor are learned by the device 10 and stored in the structured learning data 108.

The feedback information 306 stores the feedback information of the consultation result with the FP. For example, the feedback information 306 stores information such as the FP identifier 401 of the consulted FP and the impression of the consultant.

Returning to FIG. 2, the FP information 107 stores FP information, particularly information on a field of specialty and life experience. FIG. 4 is a diagram showing an example of the data structure of the FP information 107 according to the embodiment of the present invention. The FP information 107 can include the FP identifier 401, the FP basic information 402, the specialty field 403, and the life experience 404, but is not limited to these data items and can include other data items as well. is there.

The FP identifier 401 is an identifier that identifies the FP. The FP basic information 402 indicates information such as a name, affiliation, and contact information. Specialty field 403 indicates the specialty field of FP, for example, inheritance, home purchase / mortgage, financial product, and the like. The life experience 404 indicates the life experience of the FP itself, for example, the presence or absence of children, the presence or absence of own home acquisition, the presence or absence of investment experience, and the like.

Returning to FIG. 2, the structured learning data 108 stores the demographic information collected by the device 10 and the structured data of the key phrase. The structured learning data 108 may be structured data generated from key phrases aggregated for each demographic information based on the data collected from the external system 13 by the device 10. FIG. 5 is a diagram showing an image of structured data. FIG. 5 shows that the individual pieces of demographic information and each key phrase are correlated with different weights. The structured data shown in FIG. 5 is merely an example, and the attributes of the demographic information may be adopted in addition to those listed here (for example, gender, age, etc.).

(Processing flow: FP recommendation processing and machine learning processing)
FIG. 6 is a diagram illustrating the FP recommendation process and the machine learning process executed by the device 10. While referring to this processing flow, the apparatus 10 receives the information of the consultant from the consultant terminal 11, extracts the key phrase from the information, estimates the topic by referring to the structured data, and obtains the obtained information. The process of recommending an FP suitable for a topic will be described.

Further, referring to this processing flow, the process in which the device 10 machine-learns the demographic information of the consultant and the extracted key phrase as structured data will be described.

As a premise of this processing flow, it is assumed that the counselor accesses the device 10 via the counselor terminal 11 and the authentication process has already been performed.

In S601, the device 10 receives the information input by the counselor from the counselor terminal 11. The received information is stored in the consultant reception information 302. The information received from the counselor terminal 11 is, for example, information such as gender, age, presence / absence of spouse / children, residential area, occupation, annual income, financial assets held, and consultation content.

In S602, the device 10 reads the counselor information stored in the counselor reception information 302 and categorizes it into a predetermined format (for example, changing the actual age to the age, changing the residential area to the prefecture, etc.). By doing so, the demographic information of the counselor is generated and stored in the counselor demographic information 303. The predetermined format may change depending on the demographic information to be generated.

In S603, the device 10 uses two algorithms, PositionRank and EmbedRank, to read the text of the consultation content read from the consultant reception information 302 (for example, worried about the mortgage, the amount of savings is 〇〇.・ ・, Etc.) to extract key phrases. The extracted key phrases are weighted through the processing of the algorithm.

In S604, the device 10 performs machine learning by storing the demographic information of the consultant and the extracted key phrase in the structured learning data 108. The structured learning data 108 is updated with the newly stored demographic information and key phrases of the consultant. As mentioned above, each key phrase associated with the demographic information is weighted.

In S605, the device 10 reads the key phrase associated with the demographic information in the same category as the demographic information of the counselor from the structured learning data 108, and associates the read key phrase with the counselor extracted in S603. Estimate the topic by comparing it with the key phrase that is given.

As a result of comparison, the device 10 determines whether the tendency of the weight of the key phrase associated with the counselor matches the tendency of the weight of the key phrase associated with the demographic information of the same category as the demographic information of the counselor. judge. For example, in the demographic information of a man in his twenties living in Tokyo, in the case where the mortgage weight is 2 and the savings weight is 1, among the key phrases associated with the counselor, the mortgage weight is 2. If the weight of the savings is 1, it is determined that they match. If they match, the device 10 estimates the key phrase associated with the counselor as the counseling topic. Estimated topics are sorted in descending order of weight. For example, 2 mortgages, 2 savings, 1 medical expense, 1 health insurance, and so on.

On the other hand, if they do not match, the device 10 processes the data so that the different key phrases can be identified. For example, in the key phrase associated with the counselor, the weight of medical expenses is 3, but in the key phrase of the same demographic information, when the weight of medical expenses is 1, the weights of both can be distinguished. As such, the data is processed so as to be, for example, medical expenses 1 (3). That is, the weight of the demographic information in the same category can be displayed as the main display, and the weight of the key phrase derived from the sentence input by the counselor can be displayed as the sub display. Note that this display may be reversed. In another embodiment of the present invention, if they do not match, or if the difference between the two is more than a predetermined value, the key phrase associated with the counselor may not be adopted as a topic. Is.

In S606, the apparatus 10 acquires a list of FPs specializing in the field from the FP information 107 based on the topic estimated in S605. The topic can include some or all of the key phrases associated with the counselor. The device 10 sequentially selects the FP having the specialty field corresponding to the topic with the highest weight from the list (if a plurality of FPs are applicable, the FP having the specialty field corresponding to the topic with the next highest weight). Select in order). This makes it possible to determine the candidate ranking of the FP recommended to the consultant. Alternatively, the device 10 selects in order from the FP having the largest number of specialty fields corresponding to each topic. For example, if there are 5 topics, the experience value is normal, but the FP has a specialty field that can handle all 5 topics, and the experience value is excellent, but the FP has only one specialty field that can be handled. Then, the order in which the former is recommended comes first.

The device 10 notifies the FP terminal 12 in descending order of order, and receives a reply of consultation acceptance from the FP terminal 12. After that, the device 10 notifies the consultant terminal 11 of recommending the FP that has accepted the consultation. In this notification, the reason why the FP is recommended (for example, strong in the field of XX, abundant experience of △△, etc.) may be notified. When the counselor terminal 11 starts a consultation with the recommended FP, the counselor terminal 11 transmits a signal for starting the consultation to the device 10, and starts a chat with the FP terminal 12 in the chat room provided by the device 10.

Further, when the device 10 recommends the FP, the device 10 can refer to the information of the life experience 404 of the FP information 107 and preferentially recommend the FP having more points of agreement with the attribute information of the counselor. For example, when a counselor who is married and has children consults about home purchase, it is easier for an FP who is married and has children and has already purchased a home to grasp the needs of the counselor. Become.

If the recommended FP cannot immediately respond to the consultation, the device 10 can notify the consultant terminal 11 that the FP will contact the user within a predetermined time.

(Processing flow: Topic change processing)
FIG. 7 is a diagram illustrating a topic change process executed by the device 10. The process in which the device 10 changes the topic based on the content of the chat performed between the consultant and the FP will be described with reference to this process flow.

In S701, the device 10 reads out the text of the chat between the counselor and the FP. The read text may be stored in the main storage unit 102.

In S702, the device 10 newly extracts a key phrase by using two algorithms of PositionRank and EmbedRank based on the information of the sentence read in S701. The extracted new key phrases are weighted according to the content of the read text.

In S703, the device 10 performs machine learning by storing the demographic information of the consultant and the extracted new key phrase in the structured learning data 108. The structured learning data 108 is updated with the newly stored demographic information of the consultant and a new key phrase.

In S704, the device 10 presumes that the new key phrase associated with the counselor extracted in S702 is a topic that the counselor is interested in. The presumed topic was determined to have derived the potential needs of the counselor from the content of the conversation between the counselor and the FP.

In S705, the device 10 displays the topic estimated in S704 on the screen of the chat room between the counselor and the FP. FIG. 8 is a diagram showing an example of the screen 800 of the chat room. The content of the chat between the counselor and the FP, the demographic information 801 of the counselor, and the topic 802 are displayed on the screen 800. The device 10 can change the information of the topic 802 based on the topic estimated in S704. As a result, the FP can know the potential needs of the counselor from the content of the counselor's remarks, and can provide a consultation with higher user satisfaction.

The device 10 may display the topic before the change and the topic after the change in parallel, or may display only the topic after the change. Further, although topic 802 is information referred to only by the FP, it may be configured so that the consultant can refer to it.

The topic change process described with reference to FIG. 7 may be repeatedly executed at a predetermined timing (for example, when the conversation between the counselor and the FP is performed N times (N is a natural number)).

By the processing as described above, the device 10 can extract the potential needs of the counselor, thereby providing the counselor with more appropriate advice from the FP. Further, as an incidental effect, as a result of increasing the satisfaction level of the counselor with respect to the content of the consultation with the FP, the effect of increasing the proportion of users who actually meet and consult with the FP after the chat can be obtained.

The counselor can use the counselor terminal 11 to provide feedback of the consultation result with the FP to the device 10. The device 10 can store the provided feedback in the feedback information 306. Further, the FP and / or the counselor can feed back the interview information to the device 10 when the face-to-face consultation is actually conducted. The feedback interview information can be stored in the feedback information 306 by the device 10.

Although the principles of the present invention have been described above with reference to exemplary embodiments, those skilled in the art will appreciate that various embodiments that change in configuration and details can be realized without departing from the gist of the present invention. You will understand. That is, the present invention can be implemented as, for example, a system, an apparatus, a method, a program, a storage medium, or the like.

10 Device 11 Consultant terminal 12 FP terminal 13 External system 101 Control unit 102 Main storage unit 103 Auxiliary storage unit 104 Interface (IF) unit 105 Output unit 106 Consultant information 107 FP information 108 Structured learning data

Claims (11)

A matching device for matching users and experts,
User information is received from the first terminal device associated with the user and
By categorizing the received user information into a predetermined format, demographic information of the user is generated.
Based on the consultation content information included in the received user information, a first weighted key phrase is extracted using a predetermined algorithm.
A second weighted key phrase associated with demographic information in the same category as the user's demographic information is read from the trained data, and the first weighted key phrase is compared with the second weighted key phrase. And estimate the topic,
Get a list of the experts associated with the estimated topic
From the acquired list of experts, select an expert according to predetermined criteria, and select an expert.
A matching device configured to recommend the selected expert to the user. The matching device according to claim 1, wherein the matching device is further configured to provide a chat room for the user and the recommended expert to chat. The matching device is
Read the information of the chat between the user and the expert,
Based on the read chat information, a third weighted key phrase associated with the user is extracted using the predetermined algorithm.
The matching device of claim 2, further configured to display the third weighted key phrase in the chat room. The matching device is further configured to machine-learn by storing the user's demographic information and the first weighted keyphrase in the trained data and generate the updated trained data. , The matching device of claim 1. The comparison is performed by determining whether or not the tendency of each weight of the first weighted key phrase and the tendency of the weight of the second weighted key phrase match. Matching device. The matching apparatus of claim 5, further configured to presume the first weighted key phrase as the topic if it is determined to match. If it is determined that they do not match, one of the weights of the first weighted key phrase and the weight of the second weighted key phrase that do not match is set as the main display, and the other is set as the sub display. The matching device according to claim 5 , further configured as described above . The matching apparatus according to claim 1, wherein selecting the expert according to the predetermined criteria is performed based on at least information on the specialty field of the expert. The matching device according to claim 1, wherein the matching device is further configured to preferentially recommend the expert who has many matching points with the attribute information of the user. The matching device according to claim 1, wherein the predetermined algorithm is both the PositionRank and EmbedRank algorithms. A matching method performed by a matching device for user-expert matching,
Receiving user information from the first terminal device associated with the user,
By categorizing the received user information into a predetermined format, the demographic information of the user can be generated.
Extracting the first weighted key phrase by using a predetermined algorithm based on the information of the consultation content included in the received user information.
A second weighted key phrase associated with demographic information in the same category as the user's demographic information is read from the trained data, and the first weighted key phrase is compared with the second weighted key phrase. To estimate the topic and
To get a list of the experts associated with the estimated topic,
From the obtained list of experts, select experts according to predetermined criteria, and
A matching method comprising recommending the selected expert to the user.

Images