WO2019196261A1 - 一种协议文件的推送方法及终端设备 - Google Patents

一种协议文件的推送方法及终端设备 Download PDF

Info

Publication number
WO2019196261A1
WO2019196261A1 PCT/CN2018/097563 CN2018097563W WO2019196261A1 WO 2019196261 A1 WO2019196261 A1 WO 2019196261A1 CN 2018097563 W CN2018097563 W CN 2018097563W WO 2019196261 A1 WO2019196261 A1 WO 2019196261A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
protocol
candidate
target
recommendation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/097563
Other languages
English (en)
French (fr)
Inventor
童小勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019196261A1 publication Critical patent/WO2019196261A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0282—Rating or review of business operators or products
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Recommending goods or services

Definitions

  • the present application belongs to the field of data push technology, and in particular, to a method for pushing a protocol file and a terminal device.
  • Agreement documents such as travel insurance agreements, auto insurance agreements, health insurance agreements, etc.
  • users often need sales personnel to recommend before they can choose their own target agreement.
  • the recommended method of the existing agreement documents is generally based on the experience of the sales staff.
  • the screening efficiency and accuracy of the sales personnel are gradually reduced, especially for the lack of experience.
  • the salesperson cannot recommend the appropriate protocol file for the user. It can be seen that the push method of the existing protocol file has low push efficiency and low accuracy.
  • the embodiment of the present application provides a method for recommending a protocol file and a terminal device, so as to solve the problem of the push method, the push efficiency, and the low accuracy of the existing protocol file.
  • a first aspect of the embodiment of the present application provides a method for pushing a protocol file, including:
  • the embodiment of the present application obtains the user information of the target user, and imports the candidate information into the candidate protocol recommendation model based on the user information to obtain a candidate agreement that matches the target user, and calculates the score values of the multiple candidate dimensions based on the candidate protocols.
  • the recommendation coefficients of each candidate protocol are further filtered to obtain a target protocol that meets the user's requirements and is pushed to the user terminal.
  • the selection of the target protocol is no longer dependent on the experience of the salesperson, but is automatically generated by the terminal device, and the selected protocol pool is a protocol file library containing all the protocol files, and no omission occurs. The selected situation improves the accuracy of selection and the speed of selection.
  • FIG. 1 is a flowchart of an implementation of a method for pushing a protocol file according to a first embodiment of the present application
  • FIG. 2 is a flowchart of a specific implementation of a method for pushing a protocol file according to a second embodiment of the present application
  • FIG. 3 is a flowchart of a specific implementation method of a method for pushing a protocol file S105 according to a third embodiment of the present application;
  • FIG. 4 is a schematic diagram of a push recording time axis according to an embodiment of the present application.
  • FIG. 5 is a flowchart of a specific implementation method of a method for pushing a protocol file S103 according to a fourth embodiment of the present application;
  • FIG. 6 is a flowchart of a specific implementation method of a method for pushing a protocol file S101 according to a fourth embodiment of the present application;
  • FIG. 7 is a structural block diagram of a terminal device according to an embodiment of the present application.
  • FIG. 8 is a schematic diagram of a terminal device according to another embodiment of the present application.
  • the execution body of the process is a terminal device.
  • the terminal device includes but is not limited to: a mobile terminal such as a smart phone, a notebook computer, a computer, a tablet computer, and the like.
  • the terminal device may be a seat terminal, and the agent terminal pushes the selected protocol file to the user terminal of each target user.
  • FIG. 1 is a flowchart showing an implementation of a method for pushing a protocol file according to a first embodiment of the present application, which is described in detail as follows:
  • the target user selects from the user database through the terminal device, and the administrator of the terminal device may also designate a part of the user as the target user.
  • the terminal device records the selection rule of the target user, and then queries the user information of each user in the user database based on the selection rule, and selects the user that matches the selection rule as the target.
  • the selection rule may be a user whose residence is in the A area, and the terminal device queries the information of each user's residence item in the user database, and extracts the user belonging to the A area in the information as the target user.
  • Other selection rules can also be determined in the above manner, and will not be repeated here.
  • the administrator may select a part of the user as the target user from the user list based on the user list displayed by the terminal device, and the terminal device extracts the user including the list unit of the check identifier as the target user.
  • the database storing the user information may be a memory of the mobile terminal or a peripheral database server. If the user database is stored locally, the terminal device directly extracts the user information of the target user from the memory according to the user identifier of the target user; if the user database is stored in a peripheral database server, the terminal device may be based on the target user.
  • the target identifier generates an acquisition instruction of the user information, sends an acquisition instruction to the user database server, and receives the user information returned by the server.
  • the user information includes but is not limited to at least one of the following: age, address, occupation, work unit, income status, expenditure status, health status, marital status, and financial status
  • the terminal device may be based on the user information. Determine the protocol file associated with this user.
  • the user information is imported into a candidate protocol recommendation model, and a candidate protocol matching the target user is selected from the protocol file library.
  • the terminal device imports the user information into the candidate protocol recommendation model, and can determine, from the protocol file library, a protocol file that matches the user information as a candidate protocol.
  • the candidate protocol model may be a matching degree calculation function, and the terminal device calculates a matching degree between the user information and each protocol file in the protocol file library according to the candidate protocol model, and selects a matching degree that is greater than a preset matching.
  • the protocol file of the threshold is used as a candidate protocol for the target user.
  • the protocol file library can also be stored locally in the terminal device, or can be stored in a peripheral server.
  • the manner in which the terminal device acquires the candidate protocol and the user information is based on the above two methods. The method is the same, and will not be repeated here.
  • the recommendation coefficients of the respective candidate protocols are respectively calculated based on the score values of the plurality of preset dimensions of the candidate protocols.
  • each protocol file has a corresponding rating value in different preset dimensions based on the protocol type, protocol content, sales price, and rate of return of the protocol file.
  • the rating value is calculated from the above plurality of information of the protocol file.
  • the preset dimensions include, but are not limited to, a purchase quantity, a historical user rating value, a consumption matching degree, and a return rate. It can be seen that the score value of the preset dimension is not fixed, but can be relatively floating based on the feedback of the actual user and the corresponding sales situation in the process of selling the agreement file.
  • the score value of the partial dimension will be presented according to the different target users, for example, the consumption matching degree, which is calculated based on the income information of the target user, the expenditure information, and the sales amount of the agreement file.
  • the score value of each protocol file can be stored in the protocol file database, and the terminal device can query the score value of the protocol file in each dimension based on the file identifier of the protocol file; of course, the terminal device can also obtain the protocol.
  • the scoring reference data of each dimension of the file is used to calculate the scoring value of the candidate protocol in each dimension according to the scoring parameter data when determining that the protocol file is a candidate protocol.
  • the candidate protocol is determined based on whether the user information matches, but the candidate protocol, such as personality ratio, purchase amount, and the like, are not considered when selecting the candidate protocol, the number of candidate protocols is large. If all the candidate protocols are pushed to the target user, the difficulty of the user selection protocol is increased, and the purpose of accurate push cannot be achieved. Therefore, in the embodiment, in S103, the recommendation coefficient of each candidate protocol is calculated based on the score values of the respective candidate protocols in each dimension, and the target protocol that needs to be pushed to the target user is determined based on the recommendation coefficient.
  • the terminal device may add the score values of the respective dimensions, and use the obtained result as the recommendation coefficient of the candidate protocol.
  • a target protocol is selected from the candidate protocols based on the recommendation coefficient.
  • the terminal device sorts the candidate protocols according to the order of the recommended coefficients, and selects a preset number of candidate protocols as the target protocol. In other words, the terminal device may select one of the recommended coefficients as the target protocol, and may select the N candidate protocols with the largest recommended coefficient as the target protocol. Where N is a positive integer.
  • the target protocol is sent to the user terminal of the target user.
  • the terminal device after determining the target protocol, acquires the communication address of the user terminal of the target user, and initiates a connection request to the user terminal. After the communication connection is established, the terminal device connects the target protocol through the communication. Push the user terminal.
  • the terminal device may detect a current network state of the user terminal, and if the current network state meets a preset protocol push state, send the target protocol to the user terminal.
  • the terminal device may set the sending process of the target protocol to the to-be-sent state, and when detecting that the user terminal connects to the terminal device through the client, set the sending process to an active state, and The user terminal sends a target protocol.
  • the present embodiment can be applied to a push scenario of a vehicle insurance agreement.
  • the user information of the target user includes: vehicle model, purchase time, address information, and the like.
  • the terminal device is specifically a terminal device of a manual agent.
  • the manual agent selects the target user on the terminal device, and the terminal device acquires relevant information such as the vehicle model and the purchase time of the target user, and selects a vehicle insurance agreement matching the vehicle model and the vehicle service life from the vehicle insurance agreement library, that is, the candidate protocol.
  • the terminal device obtains the scores of various candidate auto insurance agreements in multiple dimensions such as sales volume, return rate, and cost performance, calculates the recommendation coefficient of each candidate auto insurance, and selects one candidate auto insurance with the highest recommendation coefficient as the target recommended to the target user.
  • the artificial agent can send the target vehicle insurance to the user terminal of the target user by means of telephone, short message, mail, link, etc., to complete the push of the auto insurance agreement.
  • the terminal device may further automatically fill in the auto insurance agreement in the auto insurance agreement based on other user information of the target user, such as the income situation and the expenditure situation, thereby reducing the user's Fill in the operation.
  • the method for pushing a protocol file obtains the user information of the target user, and imports the candidate information into the candidate protocol recommendation model based on the user information to obtain a candidate agreement matching the target user, and is based on
  • the candidate coefficients of the candidate protocols are calculated in a plurality of preset dimensions, and the recommendation coefficients of the candidate protocols are calculated, so that the candidate protocols are further filtered, and the target protocol conforming to the user requirements is obtained and pushed to the user terminal.
  • the selection of the target protocol is no longer dependent on the experience of the salesperson, but is automatically generated by the terminal device, and the selected protocol pool is a protocol file library containing all the protocol files, and no omission occurs. The selected situation improves the accuracy of the selection and the speed of selection.
  • FIG. 2 is a flowchart showing a specific implementation of a method for pushing a protocol file according to a second embodiment of the present application.
  • the method for pushing a protocol file provided by this embodiment further includes S201 to S203, which are specifically described as follows:
  • the method before the user information is imported into the candidate protocol recommendation model, and the candidate protocol matching the target user is selected from the protocol file library, the method further includes:
  • the candidate protocol recommendation model is specifically a long-term and short-term memory LSTM neural network.
  • the terminal device acquires the user information of the training user and the protocol identifier of the historical purchase agreement.
  • the number of the training users is multiple.
  • the number of the training users should be greater than 1000, thereby improving the recognition accuracy of the LSTM neural network.
  • the user database not only records the user information of each user, but also records the protocol file that the user has purchased, and the user purchases the protocol file, indicating that the protocol file is suitable for the user's use requirement, that is, the terminal.
  • the device needs to recommend the protocol file that is the same as or similar to the protocol file that the user has purchased. Therefore, the protocol file in the protocol file library that satisfies the protocol feature of the historical purchase agreement can be identified as the candidate agreement of the user. That is, the user information of the training user is used as an input reference value of the LSTM neural network, and the historical purchase protocol of the training user is used as an output reference value of the LSTM neural network, and the LSTM neural network is trained by the above two parameters.
  • the purchase record of the protocol files of all users is recorded in the user database, so the training user can be all users included in the user database, thereby eliminating the need for the administrator to generate multiple images one by one. Training the user reduces the amount of operation in the training process, and each purchase record in the user database has been actually generated, and the accuracy is high.
  • the format of the user information of each training user is the same, that is, the number of items of the user parameter items included in the user information is the same. If any training user does not collect some user parameter items during the process of collecting user information, the user information is empty, thereby ensuring that when the user information is used as an input signal when training the LSTM neural network, each channel is used. The meaning of the parameters is the same, which improves the accuracy of the LSTM neural network.
  • the learning parameters in the long-term and short-term memory LSTM neural network are adjusted such that the learning parameters satisfy the following conditions:
  • ⁇ * is the adjusted learning parameter
  • S is an identifier of the historical purchase agreement
  • I is the user information
  • I 1 , I 2 , I 3 , ..., I n are included in the user information a parameter value of each user parameter
  • n is the number of the user parameters
  • I 1 , I 2 , I 3 , . . . , I n ; ⁇ ) is when the value of the learning parameter is ⁇ , Importing the user information of the training user into the LSTM neural network, and outputting a probability value of the identifier of the historical purchase agreement of the training user
  • I 1 , I 2 , I 3 , . . . , I n ; ⁇ ) is the value of the learning parameter when the probability value takes a maximum value.
  • the LSTM neural network has N input channels, and each input channel corresponds to each user parameter included in the user information, for example, the user information includes: age, gender, annual income, average monthly expenditure, vehicle type. With 6 user parameters for the purchase date, there will be 6 input channels in the LSTM neural network. According to the number of each user parameter in the user information, the input channel corresponding to each user parameter is fixed, thus ensuring each The user parameters entered for the input channel are the same.
  • the output channel of the LSTM neural network is one for outputting an identifier of a protocol file that matches user information.
  • the number of identifiers of the output protocol files may be multiple, so the number of output channels of the LSTM neural network is one, but the number of identifiers of the output protocol files is Can be multiple.
  • the LSTM neural network includes M output channels, and the value of M is the same as the number of protocol files in the protocol file library.
  • a sequence consisting of "1" and "0" characters is output, wherein each element in the sequence corresponds to the identifier of a protocol file in the protocol file library, that is, the number of bits of the element and The numbering of the protocol files is interrelated, so the terminal device can determine which protocol files in the protocol file library are candidate protocols for the target user.
  • the LSTM neural network includes a plurality of neural layers, each of which is provided with a corresponding learning parameter, and the parameter values of the learning parameters can be adjusted to adapt to different input types and output types.
  • the learning parameter is set to a certain parameter value
  • the user information of the plurality of training users is input to the LSTM neural network, and the identifier of the protocol file is output correspondingly, and the terminal device identifies the protocol file with the purchased protocol.
  • the identifiers are compared to determine whether the current output is correct, and based on the output results of the plurality of training protocols, the probability value that the learning result is correct when the learning parameter takes the parameter value is obtained.
  • the terminal device adjusts the learning parameter so that the probability value takes a maximum value, indicating that the LSTM neural network has been adjusted.
  • the candidate protocol recommendation model is generated based on the LSTM neural network after adjusting the learning parameters.
  • the terminal device uses the LSTM neural network adjusted with the learning parameters as a candidate recommendation model, and improves the accuracy of the candidate protocol recommendation model identification.
  • the LSTM neural network is trained by the training user, and the learning parameter corresponding to the maximum output value is selected as the parameter value of the learning parameter in the LSTM neural network, thereby improving the accuracy of the candidate protocol identification. To achieve the purpose of accurate push.
  • FIG. 3 is a flowchart showing a specific implementation of a method for pushing a protocol file S105 according to the third embodiment of the present application.
  • S105 includes S1051 and S1052, and the details are as follows:
  • the target protocol is sent to the user terminal with a preset push cycle, and a push record time axis is generated; the push record time axis includes a push record node created at each push time.
  • the terminal device sends a target protocol to the user terminal of the target user in a preset push period to remind the user to subscribe to the protocol file. For example, if a user's insurance agreement is about to expire, the user needs to be reminded to renew the insurance. Therefore, the terminal device pushes the target agreement to the user in a preset push cycle, which not only provides the user with a suitable insurance agreement, but also reaches the reminder. The purpose of renewal.
  • the terminal device in the process of pushing the insurance agreement, the terminal device generates a push record time axis to record the push situation of the target user push target protocol.
  • the terminal device pushes a target protocol to the user terminal of the target user, and creates a push record node on the push record time axis according to the push time, that is, the position of the push record node on the push time axis, that is, the push time is Push the corresponding position on the timeline.
  • the push record node includes a protocol identifier and/or protocol content of the target protocol.
  • the terminal device can obtain the foregoing protocol information by clicking the push record node, so that the administrator can quickly obtain the push situation.
  • the protocol adjustment information is imported into the push record node; the protocol adjustment information is used to adjust the target protocol. Parameters.
  • the user may return a protocol adjustment information to the terminal device, where the agreement certificate information needs to be The content of the target agreement adjustment.
  • the terminal device imports the protocol adjustment information into the push node, so that the administrator can determine the modification opinion of the user.
  • the terminal device after receiving the protocol adjustment information of the user, the terminal device adjusts the target protocol, and returns the adjusted target protocol to the user terminal, so that the target user confirms the target protocol. And if the confirmation instruction based on the adjusted target protocol is received, the target protocol and the adjusted target protocol are associated, and the adjusted target protocol is identified as a target user's pending purchase agreement.
  • the terminal device may adjust the current protocol adjustment information to the previous cycle based on the adjustment information returned by the user each time.
  • the protocol adjustment information is compared, the adjustment change content is determined, and the adjustment change content is stored in the push record node, thereby facilitating the administrator to determine the change tendency of the target user's purchase tendency.
  • FIG. 4 is a schematic diagram of a push recording time axis provided by the present application.
  • the position marked by the push record node on the time axis is the push time of the push operation, and the target protocol of the push and the protocol adjustment information returned by the user are recorded in the form of a prompt box.
  • the prompt box can be popped up when the mouse is close to the node, and can also be kept popped up.
  • the administrator can select a prompt box that needs to be popped up by clicking, checking, etc., so that the attention can be compared.
  • the push record of the push time is the position marked by the push record node on the time axis.
  • the push record time axis stores the push situation of each push, thereby facilitating the administrator to quickly determine the content of each push and the feedback of the user, thereby improving the efficiency of the administrator's information acquisition, especially for the protocol. Sales people in the sales field can quickly grasp the user's purchasing tendency and improve the efficiency of sales.
  • FIG. 5 is a flowchart showing a specific implementation of a method for pushing a protocol file S103 according to the fourth embodiment of the present application.
  • the S103 includes: S1031 to S1035, which are as follows:
  • the preset dimension includes: a purchase quantity, a historical user rating value, a consumption matching degree, and a reward rate; and the rating values in the plurality of preset dimensions based on each of the candidate protocols , respectively, calculate the recommendation coefficients of each candidate protocol, including:
  • the terminal device obtains the average income amount of the target user, and determines the purchase level of the user by the average income amount. Due to different agreement documents, the amount purchased varies.
  • the terminal device can determine the purchase status of each candidate agreement by using the average income amount of the user, thereby obtaining a score value in the consumption matching degree.
  • the terminal device may query the bank account list owned by the user based on the identity information of the user, and obtain the funds flowing into the bank account within one year, thereby determining the average income of the target user. Amount.
  • the average income amount can be the monthly income amount
  • the obtained time length can be one year
  • the average monthly income of the user is estimated by the annual income
  • the length of the acquired time may be In 3 months
  • the average weekly income of the target user is estimated by the total income of three months.
  • a consumption matching degree of the target user and each of the candidate protocols is respectively calculated according to the average income amount and the purchase amount of each of the candidate protocols.
  • the terminal device determines the consumption matching degree of the target user according to the average income amount of the target user and the purchase amount of each candidate agreement. Specifically, the terminal device records a correspondence table between the consumption matching degree and the fund difference, and the terminal device calculates a difference between the average income amount of the user and the purchase amount, and queries the consumption matching degree corresponding to the difference, thereby determining the target. The degree of consumption matching between the user and the candidate agreement.
  • the terminal device may further set a consumption matching degree calculation function, import the average income amount and the purchase amount into the matching degree calculation function, and calculate a consumption matching degree between the target user and the candidate protocol; specifically, the matching degree
  • the calculation function can be:
  • Mth is the consumption matching degree
  • M income is the average income amount
  • M price is the purchase amount
  • M stand and ⁇ are the preset coefficients.
  • a return rate of each of the candidate protocols is respectively calculated according to a return amount of each of the candidate protocols and the purchase amount; and the return amount is specifically obtained by a target user when a preset condition of the candidate protocol is satisfied. The amount.
  • the terminal device obtains the return amount of the candidate agreement, that is, the amount of funds returned to the user after satisfying the condition preset in the agreement, and the purchase amount of the candidate agreement, and determines the return rate of the candidate agreement.
  • the rate of return may be the ratio between the amount of the return and the amount of the purchase.
  • the consumption matching degree and the return rate are both a ratio, that is, the above two parameters do not have corresponding dimensions. Therefore, in order to be able to perform calculation, the purchase data amount of the candidate protocol and the historical user score value need to be normalized to ensure that the above four parameters are dimensionless parameter values.
  • the process of calculating the purchase quantity normalized by the candidate protocol is specifically: acquiring the purchase quantity of each candidate agreement, and calculating the total purchase quantity accumulated by all the candidate agreements, respectively calculating the purchase quantity of each candidate agreement and the total purchase quantity.
  • the ratio between the ratios is taken as the normalized purchase quantity.
  • the terminal device matches four candidate protocols, the candidate protocol A purchase quantity is 100, the candidate protocol B purchase quantity is 50, the candidate agreement C purchase quantity is 50, and the candidate agreement D purchase quantity is 200, therefore, the above
  • the total number of purchases of the four candidate agreements is 400, so that the results of normalizing the purchase quantities of the above four candidate protocols are: candidate agreement A (0.25), candidate agreement B (0.125), candidate agreement C (0.125) ; Candidate Protocol D (0.5).
  • the process of calculating the historical user score value is specifically: obtaining a score value of each candidate for the candidate protocol, determining an average score value of the candidate agreement according to the score value of each user and the total number of users, and calculating the average score value and The ratio between the scores of the scores and the scores of the scores is the normalized historical user score. For example, if the average score of a candidate agreement is 4.3, and the score of the perfect score is 5, the historical score of the candidate protocol after normalization is 0.86.
  • the normalized purchase quantity, the normalized historical user score value, the consumption matching degree, and the return rate are imported into the recommended coefficient calculation model to obtain the recommendation coefficients of each of the candidate protocols;
  • the recommendation coefficient calculation model is specifically:
  • Q is the recommendation coefficient
  • SL' is the normalized purchase quantity
  • Pnt' is the normalized historical user rating value
  • Mth is the consumption matching degree
  • PB is the return Rate
  • ⁇ is the preset compensation coefficient.
  • the terminal device imports the above parameters into the recommendation coefficient calculation model to determine the recommendation coefficient of the candidate protocol. Since the larger the number of purchases, the more the candidate agreement is welcomed by the user, the recommendation coefficient is positively related to the number of purchases. Similarly, the higher the value of the historical score, the consumption match, and the rate of return, the more suitable the user is to purchase, and therefore the positive correlation with the recommendation coefficient. It should be noted that min(SL', Pnt', Mth, PB) specifically takes the minimum value of the parameter values among the four parameters.
  • the recommendation coefficients of each candidate protocol are calculated through four dimensions, thereby improving the accuracy of the recommendation coefficient, and achieving the purpose of accurately pushing the target protocol for the target user.
  • FIG. 6 is a flowchart showing a specific implementation of a method for pushing a protocol file S101 according to the fifth embodiment of the present application.
  • the method for pushing a protocol file provided by this embodiment provides the user information of the target user, including: S1011 to S1013, and the details are as follows:
  • the terminal device obtains the effective date of each user's purchased agreement in the user database, and determines whether the purchased agreement of each user is invalid or ready to expire by the effective date, thereby promptly reminding the user to re-purchase the agreement.
  • the terminal device in order to determine the remaining effective duration of each user's purchased agreement, in addition to obtaining the effective date, the terminal device also acquires the current date, and calculates a difference between the current date and the effective date as the purchased The remaining effective duration of the agreement.
  • the user corresponding to the purchased protocol whose remaining effective duration is less than the preset duration threshold is selected as the target user.
  • the remaining valid duration is greater than or equal to the preset duration threshold, it indicates that the user's protocol expires after a long time, so there is no need to remind the user to re-purchase the agreement, thus identifying the user as a non-target
  • the remaining effective duration is less than the preset duration threshold, it indicates that the user's protocol is about to expire and needs to be re-purchased, so the user is identified as the target user, and the appropriate target protocol is pushed to the target user.
  • the valid validity date of each purchased agreement is obtained, the remaining valid duration of each purchased agreement is determined, and the user of the purchased agreement whose remaining effective duration is less than the preset duration threshold is selected as the target user, thereby Avoid unnecessary economic losses caused by users forgetting to purchase agreements.
  • FIG. 7 is a structural block diagram of a terminal device according to an embodiment of the present application, where each unit included in the terminal device is used to execute each step in the embodiment corresponding to FIG. 1.
  • each unit included in the terminal device is used to execute each step in the embodiment corresponding to FIG. 1.
  • please refer to the related description in the embodiment corresponding to FIG. 1 and FIG. For the convenience of explanation, only the parts related to the present embodiment are shown.
  • the terminal device includes:
  • the user information obtaining unit 71 is configured to acquire user information of the target user.
  • a candidate protocol selection unit 72 configured to import the user information into a candidate protocol recommendation model, and select a candidate protocol that matches the target user from the protocol file library;
  • a recommendation coefficient calculation unit 73 configured to separately calculate recommendation coefficients of each candidate protocol based on score values of the plurality of preset dimensions of each of the candidate protocols
  • the target protocol determining unit 74 is configured to select a target protocol from the candidate protocols based on the recommended coefficient
  • the target protocol pushing unit 75 is configured to send the target protocol to the user terminal of the target user.
  • the terminal device further includes:
  • a training data acquiring unit configured to acquire user information of the training user and an identifier of the historical purchase agreement
  • a learning parameter training unit configured to adjust a learning parameter in the long-term and short-term memory LSTM neural network based on the user information and the identifier of the historical purchase protocol, so that the learning parameter satisfies the following conditions:
  • ⁇ * is the adjusted learning parameter
  • S is an identifier of the historical purchase agreement
  • I is the user information
  • I 1 , I 2 , I 3 , ..., I n are included in the user information a parameter value of each user parameter
  • n is the number of the user parameters
  • I 1 , I 2 , I 3 , . . . , I n ; ⁇ ) is when the value of the learning parameter is ⁇ , Importing the user information of the training user into the LSTM neural network, and outputting a probability value of the identifier of the historical purchase agreement of the training user
  • I 1 , I 2 , I 3 , . . . , I n ; ⁇ ) is a value of the learning parameter when the probability value takes a maximum value
  • the candidate protocol recommendation model generating unit is configured to generate the candidate protocol recommendation model based on the LSTM neural network after adjusting the learning parameters.
  • the target protocol pushing unit 75 includes:
  • a push record node creating unit configured to send the target protocol to the user terminal in a preset push cycle, and generate a push record time axis; the push record time axis includes a push record node created at each push time;
  • a protocol adjustment information importing unit configured to import the protocol adjustment information into the push record node if receiving the protocol adjustment information returned by the user terminal according to the target protocol; and the protocol adjustment information is used for adjusting The parameters of the target protocol.
  • the preset dimension includes: a purchase quantity, a historical user rating value, a consumption matching degree, and a reward rate;
  • the recommendation coefficient calculation unit 73 includes:
  • An average income obtaining unit configured to obtain an average income amount of the target user
  • a consumption matching degree calculating unit configured to calculate, according to the average income amount and the purchase amount of each of the candidate agreements, a consumption matching degree of the target user and each of the candidate protocols;
  • a rate of return calculation unit configured to calculate a rate of return of each of the candidate protocols according to a return amount of each of the candidate agreements and the purchase amount
  • a normalization unit configured to normalize the purchase quantity of the candidate agreement and the historical user rating value
  • a parameter importing calculation unit configured to import a normalized purchase quantity, a normalized historical user score value, a consumption matching degree, and a return rate into a recommended coefficient calculation model to obtain each candidate agreement Recommendation coefficient;
  • the recommendation coefficient calculation model is specifically:
  • Q is the recommendation coefficient
  • SL' is the normalized purchase quantity
  • Pnt' is the normalized historical user rating value
  • Mth is the consumption matching degree
  • PB is the return Rate
  • ⁇ is the preset compensation coefficient.
  • the user information acquiring unit 71 includes:
  • An effective date obtaining unit configured to obtain an effective date of a purchased agreement of each user in the user database
  • a remaining effective duration calculating unit configured to separately calculate remaining remaining durations of each of the purchased agreements according to the valid date and the current date
  • the target user determining unit is configured to select, as the target user, a user corresponding to the purchased protocol whose remaining effective duration is less than a preset duration threshold.
  • the selection of the target protocol of the terminal device is no longer dependent on the experience of the salesperson, but is automatically generated by the terminal device, and the selected protocol pool is a protocol file library containing all protocol files, and no omission occurs.
  • the selected situation improves the accuracy of selection and the speed of selection.
  • FIG. 8 is a schematic diagram of a terminal device according to another embodiment of the present application.
  • the terminal device 8 of this embodiment includes a processor 80, a memory 81, and computer readable instructions 82 stored in the memory 81 and operable on the processor 80, such as protocol files. Push the program.
  • the processor 80 executes the computer readable instructions 82, the steps in the embodiment of the push method of the respective protocol files are implemented, such as S101 to S105 shown in FIG. 1.
  • the processor 80 executes the computer readable instructions 82, the functions of the various units in the various apparatus embodiments described above are implemented, such as the functions of the modules 71 to 75 shown in FIG.
  • the computer readable instructions 82 may be partitioned into one or more units, the one or more units being stored in the memory 81 and executed by the processor 80 to complete the application.
  • the one or more units may be a series of computer readable instruction instructions segments capable of performing a particular function for describing the execution of the computer readable instructions 82 in the terminal device 8.
  • the computer readable instructions 82 may be segmented into a user information acquisition unit, a candidate protocol selection unit, a recommendation coefficient calculation unit, a target protocol determination unit, and a target protocol push unit, each unit having a specific function as described above.
  • the terminal device 8 may be a computing device such as a desktop computer, a notebook, a palmtop computer, or a cloud server.
  • the terminal device may include, but is not limited to, a processor 80 and a memory 81. It will be understood by those skilled in the art that FIG. 8 is merely an example of the terminal device 8, and does not constitute a limitation on the terminal device 8, and may include more or less components than those illustrated, or combine some components, or different components.
  • the terminal device may further include an input/output device, a network access device, a bus, and the like.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Technology Law (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请适用于数据推送技术领域，提供了一种协议文件的推送方法及终端设备，包括：获取目标用户的用户信息；将用户信息导入候选协议推荐模型，从协议文件库中选取与目标用户匹配的候选协议；基于各个候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；基于推荐系数，从候选协议中选取目标协议；将目标协议发送给目标用户的用户终端。本申请中，目标协议的选取不再依赖销售人员的经验，而是通过终端设备自动生成，并且选取的协议池是包含所有协议文件的协议文件库，不会出现遗漏选取的情况，提高了选取的准确率以及选取速度。

Description

一种协议文件的推送方法及终端设备
本申请申明享有2018年04月08日递交的申请号为201810305550.2、名称为“一种协议文件的推送方法及终端设备”中国专利申请的优先权，该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本申请属于数据推送技术领域，尤其涉及一种协议文件的推送方法及终端设备。
背景技术
协议文件，例如出行保险协议、汽车保险协议、健康保险协议等，由于种类繁多，条款各异，用户常常需要销售人员进行推荐后才能选择到合适自己的目标协议。然而现有的协议文件的推荐方法，一般是根据销售人员的经验进行筛选，随着协议文件数量的不断增多以及种类的不断递增，销售人员的筛选效率以及准确度也逐步降低，特别对于经验不足的销售人员，无法为用户推荐合适的协议文件。可见，现有的协议文件的推送方法，推送效率以及准确率较低。
技术问题
有鉴于此，本申请实施例提供了一种协议文件的推荐方法及终端设备，以解决现有的协议文件的推送方法，推送效率以及准确率较低的问题。
技术解决方案
本申请实施例的第一方面提供了一种协议文件的推送方法，包括：
获取目标用户的用户信息；
将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议；
基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
基于所述推荐系数，从候选协议中选取目标协议；
将所述目标协议发送给所述目标用户的用户终端。
有益效果
本申请实施例通过获取目标用户的用户信息，并基于该用户信息导入到候选协议推荐模型，得到与该目标用户匹配的候选协议，并基于各个候选协议在多个预设维度的评分值，计算各个候选协议的推荐系数，从而对候选协议进行进一步筛选，得到符合用户需求的目标协议，并推送给用户终端。与现有的协议文件推送方法相比，目标协议的选取不再依赖 销售人员的经验，而是通过终端设备自动生成，并且选取的协议池是包含所有协议文件的协议文件库，不会出现遗漏选取的情况，提高了选取的准确率以及选取速度。
附图说明
图1是本申请第一实施例提供的一种协议文件的推送方法的实现流程图；
图2是本申请第二实施例提供的一种协议文件的推送方法具体实现流程图；
图3是本申请第三实施例提供的一种协议文件的推送方法S105具体实现流程图；
图4是本申请一实施例提供的一种推送记录时间轴的示意图；
图5是本申请第四实施例提供的一种协议文件的推送方法S103具体实现流程图；
图6是本申请第四实施例提供的一种协议文件的推送方法S101具体实现流程图；
图7是本申请一实施例提供的一种终端设备的结构框图；
图8是本申请另一实施例提供的一种终端设备的示意图。
本发明的实施方式
在本申请实施例中，流程的执行主体为终端设备。该终端设备包括但不限于：智能手机、笔记本电脑、计算机、平板电脑等移动终端。特别地，该终端设备可以为一坐席终端，坐席终端将选取得到的协议文件推送给各个目标用户的用户终端。图1示出了本申请第一实施例提供的协议文件的推送方法的实现流程图，详述如下：
在S101中，获取目标用户的用户信息。
在本实施例中，目标用户通过终端设备从用户数据库中进行选取，也可以为终端设备的管理员指定部分用户为目标用户。若目标用户为终端设备自行选取，在该情况下，终端设备记录了目标用户的选取规则，然后基于该选取规则查询用户数据库中各个用户的用户信息，选取与该选取规则相匹配的用户作为目标用户。举例性地，该选取规则可以为居住地为A地区的用户，则终端设备查询用户数据库中各个用户居住地项目的信息，提取该信息中属于A地区的用户作为目标用户。其他选取规则同样可以通过上述方式进行确定，在此不再一一赘述。
在本实施例中，管理员可以基于终端设备显示的用户列表，从用户列表中勾选部分用户作为目标用户，终端设备提取包含勾选标识的列表单元的用户作为目标用户。
在本实施例中，存储用户信息的数据库可以为移动终端的存储器，也可以为一外设数据库服务器。若用户数据库存储于本地，则终端设备直接根据目标用户的用户标识从存储器中提取该目标用户的用户信息即可；若该用户数据库存储于一外设数据库服务器，则终端设备可以根据该目标用户的目标标识生成一个用户信息的获取指令，向该用户数据库服 务器发送一个获取指令，并接收该服务器返回的用户信息。
在本实施例中，用户信息包括但不限于以下至少一种：年龄、住址、职业、工作单位、收入情况、支出情况、健康情况、婚姻状况以及财务状况等信息，终端设备可以基于上述用户信息确定与该用户相关的协议文件。
在S102中，将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议。
在本实施例中，终端设备在获取了用户信息后，将用户信息导入到候选协议推荐模型，即可从协议文件库中确定得到与该用户信息相匹配的协议文件作为候选协议。可选地，该候选协议模型可以为一匹配度计算函数，终端设备将根据所述候选协议模型计算该用户信息与协议文件库中各个协议文件的匹配度，并选取匹配度大于预设的匹配阈值的协议文件作为该目标用户的候选协议。
在本实施例中，与用户数据库类似，协议文件库同样也可以存储于终端设备本地，也可以存储一外设服务器中，其中基于上述两种方式，终端设备获取候选协议的方式与用户信息的方式相同，在此不再一一赘述。
在S103中，基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数。
在本实施例中，每个协议文件基于协议文件的协议类型、协议内容、销售价格以及回报率，在不同的预设维度有相应的评分值。该评分值是通过协议文件的上述多项信息计算得到的。其中，该预设维度包括但不限于：购买数量、历史用户评分值、消费匹配度以及回报率。由此可见，预设维度的评分值并非固定不变的，而是可以在协议文件在出售的过程中，基于实际用户的反馈以及对应的销售情况会相对浮动。并且，部分维度的评分值会根据目标用户的不同呈现不同的分数，例如消费匹配度，即会基于目标用户的收入信息、支出信息以及协议文件的销售金额计算得到。需要说明的是，每一个协议文件的评分值可以存储于协议文件数据库中，终端设备基于协议文件的文件标识即可查询得到该协议文件在各个维度的评分值；当然，终端设备还可以获取协议文件在各个维度的评分参考数据，在确定一协议文件为候选协议时则根据评分参数数据计算该候选协议在各个维度的评分值。
在本实施例中，由于候选协议是基于与用户信息的是否匹配而确定的，但候选协议例如性格比、购买量等因素在选取候选协议时并非考虑，因此候选协议的数量较多。若将所有候选协议均推送给目标用户，则会增加用户选择协议的难度，无法实现精准推送的目的。因此，本实施例在S103中，基于各个候选协议在各个维度的评分值，计算每一个候选协议的推荐系数，并基于推荐系数确定需要对目标用户进行推送的目标协议。
在本实施例中，终端设备可以将各个维度的评分值进行相加，将相加后得到的结果作为该候选协议的推荐系数。
在S104中，基于所述推荐系数，从候选协议中选取目标协议。
在本实施例中，终端设备在计算了各个候选协议的推荐系数后，基于推荐系数从大到小的次序对候选协议进行排序，并从中选取预设数量的候选协议作为目标协议。换而言之，终端设备可以选取推荐系数最大的一个作为目标协议，可以选取推荐系数最大的N个候选协议作为目标协议。其中，N为正整数。
在S105中，将所述目标协议发送给所述目标用户的用户终端。
在本实施例中，终端设备在确定了目标协议后，获取目标用户的用户终端的通信地址，并向该用户终端发起连接请求，当通信连接建立后，终端设备会将目标协议通过该通信连接推送该用户终端。可选地，终端设备可以检测用户终端当前的网络状态，若当前的网络状态满足预设的协议推送状态，则向该用户终端发送目标协议。
在本实施例中，终端设备可以将目标协议的发送进程设置为待发送状态，当检测到用户终端通过客户端与该终端设备进行连接时，则将该发送进程设置为激活状态，并向该用户终端发送目标协议。
特别地，本实施例可以应用于车辆保险协议的推送场景。在该情况下，目标用户的用户信息包括：车辆型号、购车时间、地址信息等相关信息。终端设备具体为人工坐席的终端设备。人工坐席在终端设备上选择目标用户，终端设备会获取该目标用户的车辆型号、购车时间等相关信息，从车险协议库中选取与该车辆型号匹配以及与车辆使用年限相符的车险协议，即候选协议。然后终端设备会获取各个候选的车险协议在销售量、回报率、性价比等多个维度的评分值，计算各个候选车险的推荐系数，并选取推荐系数最高的一个候选车险作为推荐给目标用户的目标车险。人工坐席可以通过电话、短信、邮件、链接等方式将该目标车险发送给目标用户的用户终端，完成车险协议的推送的目的。优选地，终端设别在推送目标协议给用户终端前，还可以基于该目标用户的其他用户信息，如收入情况以及支出情况，将车险协议中需要用户自定义部分自动填写，从而减少了用户的填写操作。
以上可以看出，本申请实施例提供的一种协议文件的推送方法通过获取目标用户的用户信息，并基于该用户信息导入到候选协议推荐模型，得到与该目标用户匹配的候选协议，并基于各个候选协议在多个预设维度的评分值，计算各个候选协议的推荐系数，从而对候选协议进行进一步筛选，得到符合用户需求的目标协议，并推送给用户终端。与现有的协议文件推送方法相比，目标协议的选取不再依赖销售人员的经验，而是通过终端设备自动生成，并且选取的协议池是包含所有协议文件的协议文件库，不会出现遗漏选取的情况， 提高了选取的准确率以及选取速度。
图2示出了本申请第二实施例提供的一种协议文件的推送方法的具体实现流程图。参见图2所示，相对于图1所述实施例，本实施例提供的一种协议文件的推送方法中还包括S201～S203，具体详述如下：
进一步地，作为本申请另一实施例，在将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议之前，还包括：
在S201中，获取训练用户的用户信息和历史购买协议的标识。
在本实施例中，候选协议推荐模型具体为一长短期记忆LSTM神经网络，为了提高该LSTM神经网络输出候选协议的正确率，需要输入训练数据对该神经网络进行学习训练。因此，在S201中，终端设备会获取训练用户的用户信息以及历史购买协议的协议标识。其中，训练用户的个数为多个，优选地，该训练用户的个数应大于1000个，从而提高该LSTM神经网络的识别准确性。
在本实施例中，用户数据库中不仅记录了各个用户的用户信息，还记录了该用户曾经购买过的协议文件，而用户购买该协议文件则表明该协议文件适合该用户的使用需求，即终端设备需要推荐与用户曾经购买的协议文件相同或相近的协议文件，因此可将协议文件库中满足历史购买协议的协议特征的协议文件，识别为该用户的候选协议。即把训练用户的用户信息作为LSTM神经网络的输入参考值，把训练用户的历史购买协议作为LSTM神经网络的输出参考值，通过上述两个参数对LSTM神经网络进行训练。
优选地，在本实施例中，如上所述，用户数据库中记载了所有用户的协议文件的购买记录，因此训练用户可以为用户数据库中包含的所有用户，从而无需管理员一一模拟生成多个训练用户，减少了训练过程的操作量，而且用户数据库中各个购买记录均是已经实际产生的，准确度较高。
需要说明的是，各个训练用户的用户信息的格式是相同的，即用户信息中包含的用户参数项的项数是相同的。若任一训练用户在用户信息采集的过程中部分用户参数项并未采集得到，则该用户信息为空，从而保证了在对LSTM神经网络进行训练时，将用户信息作为输入信号时，各个通道的参数含义是相同的，提高了LSTM神经网络的准确性。
在S202中，基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
Figure PCTCN2018097563-appb-000001
其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息； I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值。
在本实施例中，LSTM神经网络有N个输入通道，每个输入通道对应于用户信息中包含的各项用户参数，例如用户信息中包含：年龄、性别、年收入、平均月支出、车辆类型、购车日期6个用户参数，则LSTM神经网络中则会设置有6个输入通道，依据各个用户参数在用户信息中的编号，为每个用户参数固定与之对应的输入通道，从而保证每个输入通道输入的用户参数是相同的。该LSTM神经网络的输出通道为一个，用于输出与用户信息相匹配的协议文件的标识。需要说明的是，若用户购买了多个协议文件，则输出的协议文件的标识数量可以为多个，因此LSTM神经网络的输出通道的个数为1个，但输出的协议文件的标识的数量可以为多个。
优选地，LSTM神经网络包含M个输出通道，M的值与协议文件库中协议文件的个数相同。用户信息通过该LSTM神经网络后，会输出以“1”和“0”字符构成的序列，其中，该序列中各个元素对应于协议文件库中的一个协议文件的标识，即元素的位数与协议文件的编号是相互关联的，因此，终端设备能够确定协议文件库中哪些协议文件为目标用户的候选协议。
在本实施例中，LSTM神经网络中包含多个神经层，每个神经层设置有相应的学习参数，通过调整学习参数的参数值能够适应不同输入类型以及输出类型。当学习参数设置为某一参数值时，将多个训练用户的用户信息输入到该LSTM神经网络，将对应输出一系列的协议文件的标识，终端设备会将协议文件的标识与已购买协议的标识进行比对，确定本次输出是否正确，并且基于多个训练协议的输出结果，得到该学习参数取该参数值时输出结果正确的概率值。终端设备会调整该学习参数，以使该概率值取最大值，则表示该LSTM神经网络已经调整完毕。
在S203中，基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
在本实施例中，终端设备将调整了学习参数后的LSTM神经网络作为候选推荐模型，提高了候选协议推荐模型识别的准确率。
在本申请实施例中，通过训练用户对LSTM神经网络进行训练，选取输出结果正确的概率值最大时对应的学习参数作为LSTM神经网络中学习参数的参数值，从而提高了候选 协议识别的准确性，实现精准推送的目的。
图3示出了本申请第三实施例提供的一种协议文件的推送方法S105的具体实现流程图。参见图3所示，相对于图1所述实施例，本实施例提供的一种协议文件的推送方法中S105包括S1051以及S1052，具体详述如下：
在S1051中，以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点。
在本实施例中，终端设备会以预设的推送周期向目标用户的用户终端发送目标协议，以提醒用户认购协议文件。例如，某一用户的保险协议将要到期，需要提醒用户续保，因此终端设备会以预设的推送周期，向该用户推送目标协议，不仅能够为用户提供合适的保险协议，还能够达到提醒续保的目的。
在本实施例中，终端设备在推送保险协议的过程中，会生成一推送记录时间轴，来记录对目标用户推送目标协议的推送情况。其中，终端设备每向目标用户的用户终端推送一次目标协议，将根据推送时刻在该推送记录时间轴上创建一个推送记录节点，即推送记录节点在推送时间轴上的位置，即为推送时刻在推送时间轴上对应的位置。
优选地，该推送记录节点中包含该目标协议的协议标识和/或协议内容。终端设备可通过点击该推送记录节点，获取上述协议信息，方便管理员快速获取推送情况。
在S1052中，若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
在本实施例中，用户若接收到终端设备发送的目标协议后，对该目标协议中部分自定义选项进行变更，则可向终端设备返回一个协议调整信息，该协议铁证信息中包含了需要对目标协议进行调整的内容。终端设备在接收到协议调整信息后，会将该协议调整信息导入到推送节点内，以便管理员确定该用户的修改意见。
可选地，在本实施例中，终端设备在接收到用户的协议调整信息后，会对目标协议进行调整，并将调整后的目标协议返回给用户终端以便目标用户对该目标协议进行确认。若接收到基于该调整后的目标协议的确认指令，则关联该目标协议以及该调整后的目标协议，识别该调整后的目标协议为目标用户的待购买协议。
可选地，若每个推送周期推送的目标协议的标识相同，且每一次用户均反馈协议调整信息，则终端设备可以基于用户每一次返回的调整信息，将当前的协议调整信息与上一周期的协议调整信息进行比对，确定调整变更内容，并将调整变更内容存储于推送记录节点，从而方便管理员确定目标用户的购买倾向的变更情况。
举例性地，图4为本申请提供的一种推送记录时间轴的示意图。如图4所示，每个推送记录节点在该时间轴所标记的位置为该次推送操作的推送时刻，并且通过提示框的形式，记录该次推送的目标协议以及用户返回的协议调整信息。当然，该提示框可以在管理在鼠标靠近该节点时才弹出，也可以一直保持弹出状态，优选地，管理员可以通过点击、勾选等方式选取需要弹出的提示框，从而能够比对关注的推送时刻的推送记录情况。
在本申请实施例中，通过推送记录时间轴存储每一次推送的推送情况，从而方便管理员快速确定每一次推送的内容以及用户的反馈意见，提高了管理员的信息获取的效率，特别对于协议销售领域的销售人员，能够快速掌握用户的购买倾向，提高了销售的效率。
图5示出了本申请第四实施例提供的一种协议文件的推送方法S103的具体实现流程图。参见图5所示，相对于图1所述实施例，本实施例提供的一种协议文件的推送方法中所述S103包括：S1031～S1035，具体详述如下：
进一步地，作为本申请另一实施例，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数，包括：
在S1031中，获取所述目标用户的平均收入金额。
在本实施例中，终端设备会获取目标用户的平均收入金额，通过该平均收入金额确定该用户的购买水平。由于不同的协议文件，购买的金额各不相同。终端设备可通过该用户的平均收入金额，确定购买各个候选协议的购买情况，从而得到在消费匹配度的评分值。
在本实施例中，终端设备可以基于该用户的身份信息，查询该用户拥有的银行账户列表，并获取在一年时间内，流进该银行账户的资金情况，从而确定该目标用户的平均收入金额。当然，该平均收入金额可以为月收入金额，则获取的时间长度可以为一年，通过年收入估算该用户的平均月收入；若该平均收入金额为周收入金额，则获取的时间长度可以为3个月，通过三个月的总收入情况估算该目标用户的平均周收入。
在S1032中，根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度。
在本实施例中，终端设备会根据该目标用户的平均收入金额与各个候选协议的购买金额，确定该目标用户的消费匹配度。具体地，终端设备记载有以消费匹配度与资金差额的对应关系表，终端设备计算用户的平均收入金额与购买金额的差值，查询该差值所对应的消费匹配度，从而确定了该目标用户与候选协议之间的消费匹配度。
当然，终端设备还可以设置一消费匹配度计算函数，将平均收入金额与购买金额导入到该匹配度计算函数中，计算目标用户与该候选协议之间的消费匹配度；具体地，该匹配 度计算函数可以为：
Figure PCTCN2018097563-appb-000002
其中，Mth为消费匹配度；M income为平均收入金额；M price为购买金额；M stand以及ε为预设系数。
在S1033中，根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；所述回报金额具体为满足所述候选协议的预设条件时目标用户可获取的金额。
在本实施例中，终端设备会获取该候选协议的回报金额，即满足协议中预设的条件则返还用户的资金金额，以及该候选协议的购买金额，确定该候选协议的回报率。具体地，该回报率可以为回报金额与购买金额之间的比值。
在S1034中，对所述候选协议的购买数量以及历史用户评分值进行归一化处理。
在本实施例中，由于消费匹配度以及回报率均是一比值，即上述两个参数均不存在对应的量纲。因此为了能够进行计算，需要将候选协议的购买数据量以及历史用户评分值均进行归一化处理，从而保证上述四个参数均是无量纲的参数值。
具体地，计算候选协议归一化的购买数量的过程具体为：获取各个候选协议的购买数量，并计算所有候选协议累计的总购买数量，分别计算各个候选协议的购买数量与该总购买数量之间的比值，将该比值作为归一化后的购买数量。例如，终端设备匹配得到4个候选协议，候选协议A的购买数量为100，候选协议B的购买数量为50，候选协议C的购买数量为50，候选协议D的购买数量为200，因此，上述4个候选协议的总购买数量为400，从而对上述四个候选协议的购买数量进行归一化处理的结果为：候选协议A(0.25)，候选协议B(0.125)，候选协议C(0.125)；候选协议D(0.5)。
具体地，计算历史用户评分值的过程具体为：获取各个用户对该候选协议的评分值，根据各个用户的评分值以及用户总数，确定该候选协议的平均评分值，并计算该平均评分值与评分值满分的分值之间的比值，得到归一化后的历史用户评分值。例如，某一候选协议的平均评分值为4.3分，而满分的分值为5分，则该候选协议归一化后的历史评分值为0.86。
在S1035中，将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
Figure PCTCN2018097563-appb-000003
其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
在本实施例中，终端设备在计算了候选协议的四个评分值之后，将上述参数导入到推荐系数计算模型中，确定该候选协议的推荐系数。由于购买数量越大，则表示该候选协议越受用户欢迎，因此推荐系数与购买数量成正相关。同样地，历史评分值、消费匹配度以及回报率的数值越高，则越适合该用户购买，因此也与推荐系数成正相关。需要说明的是，min(SL′,Pnt′,Mth,PB)具体为取四个参数中参数值最小值。
在本申请实施例中，通过四个维度计算各个候选协议的推荐系数，从而提高了推荐系数的准确率，实现为目标用户精准推送目标协议的目的。
图6示出了本申请第五实施例提供的一种协议文件的推送方法S101的具体实现流程图。参见图6所示，相对于图1-图4所述实施例，本实施例提供的一种协议文件的推送方法所述获取目标用户的用户信息，包括：S1011～S1013，具体详述如下：
在S1011中，获取用户数据库中各个用户的已购买协议的有效日期。
在本实施例中，终端设备会获取用户数据库中每个用户已购买协议的有效日期，通过有效日期确定各个用户的已购买协议是否失效或准备失效，从而能够及时提醒用户重新购买协议。
在S1012中，根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长。
在本实施例中，终端设备为了确定各个用户的已购买协议的剩余有效时长，除了获取有效日期外，还会获取当前日期，并计算当前日期与有效日期之间的差值，作为该已购买协议的剩余有效时长。
在S1013中，选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
在本实施例中，若该剩余有效时长大于或等于预设的时长阈值，则表示该用户的协议在较长时间之后才到期，因此无需提醒用户重新购买协议，因此识别该用户为非目标用户；若该剩余有效时长小于预设的时长阈值，则表示该用户的协议快要失效，需要重新购买，因此会将该类用户识别为目标用户，并向该目标用户推送合适的目标协议。
在本申请实施例中，通过获取各个用户的已购买协议的有效日期，确定各个已购买协议的剩余有效时长，并选取剩余有效时长小于预设时长阈值的已购买协议的用户作为目标 用户，从而避免用户因忘记购买协议而造成不必要的经济损失。
应理解，上述实施例中各步骤的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
图7示出了本申请一实施例提供的一种终端设备的结构框图，该终端设备包括的各单元用于执行图1对应的实施例中的各步骤。具体请参阅图1与图1所对应的实施例中的相关描述。为了便于说明，仅示出了与本实施例相关的部分。
参见图7，所述终端设备包括：
用户信息获取单元71，用于获取目标用户的用户信息；
候选协议选取单元72，用于将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议；
推荐系数计算单元73，用于基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
目标协议确定单元74，用于基于所述推荐系数，从候选协议中选取目标协议；
目标协议推送单元75，用于将所述目标协议发送给所述目标用户的用户终端。
可选地，终端设备还包括：
训练数据获取单元，用于获取训练用户的用户信息和历史购买协议的标识；
学习参数训练单元，用于基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
Figure PCTCN2018097563-appb-000004
其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息；I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值；
候选协议推荐模型生成单元，用于基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
可选地，目标协议推送单元75包括：
推送记录节点创建单元，用于以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点；
协议调整信息导入单元，用于若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
可选地，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述推荐系数计算单元73包括：
平均收入获取单元，用于获取所述目标用户的平均收入金额；
消费匹配度计算单元，用于根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度；
回报率计算单元，用于根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；
归一化单元，用于对所述候选协议的购买数量以及历史用户评分值进行归一化处理；
参数导入计算单元，用于将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
Figure PCTCN2018097563-appb-000005
其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
可选地，用户信息获取单元71包括：
有效日期获取单元，用于获取用户数据库中各个用户的已购买协议的有效日期；
剩余有效时长计算单元，用于根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长；
目标用户确定单元，用于选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
因此，本申请实施例提供的终端设备同样目标协议的选取不再依赖销售人员的经验，而是通过终端设备自动生成，并且选取的协议池是包含所有协议文件的协议文件库，不会出现遗漏选取的情况，提高了选取的准确率以及选取速度。
图8是本申请另一实施例提供的一种终端设备的示意图。如图8所示，该实施例的终端设备8包括：处理器80、存储器81以及存储在所述存储器81中并可在所述处理器80上运行的计算机可读指令82，例如协议文件的推送程序。所述处理器80执行所述计算机可读指令82时实现上述各个协议文件的推送方法实施例中的步骤，例如图1所示的S101 至S105。或者，所述处理器80执行所述计算机可读指令82时实现上述各装置实施例中各单元的功能，例如图7所示模块71至75功能。
示例性的，所述计算机可读指令82可以被分割成一个或多个单元，所述一个或者多个单元被存储在所述存储器81中，并由所述处理器80执行，以完成本申请。所述一个或多个单元可以是能够完成特定功能的一系列计算机可读指令指令段，该指令段用于描述所述计算机可读指令82在所述终端设备8中的执行过程。例如，所述计算机可读指令82可以被分割成用户信息获取单元、候选协议选取单元、推荐系数计算单元、目标协议确定单元以及目标协议推送单元，各单元具体功能如上所述。
所述终端设备8可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述终端设备可包括，但不仅限于，处理器80、存储器81。本领域技术人员可以理解，图8仅仅是终端设备8的示例，并不构成对终端设备8的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件，例如所述终端设备还可以包括输入输出设备、网络接入设备、总线等。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。

Claims (20)

  1. 一种协议文件的推送方法，其特征在于，包括：
    获取目标用户的用户信息；
    将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议；
    基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
    基于所述推荐系数，从候选协议中选取目标协议；
    将所述目标协议发送给所述目标用户的用户终端。
  2. 根据权利要求1所述的推送方法，其特征在于，在将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议之前，还包括：
    获取训练用户的用户信息和历史购买协议的标识；
    基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
    Figure PCTCN2018097563-appb-100001
    其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息；I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值；
    基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
  3. 根据权利要求1所述的推送方法，其特征在于，所述将所述目标协议发送给所述目标用户的用户终端，包括：
    以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点；
    若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
  4. 根据权利要求1所述的推送方法，其特征在于，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数，包括：
    获取所述目标用户的平均收入金额；
    根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度；
    根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；所述回报金额具体为满足所述候选协议的预设条件时目标用户可获取的金额；
    对所述候选协议的购买数量以及历史用户评分值进行归一化处理；
    将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
    Figure PCTCN2018097563-appb-100002
    其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
  5. 根据权利要求1-4任一项所述的推送方法，其特征在于，所述获取目标用户的用户信息，包括：
    获取用户数据库中各个用户的已购买协议的有效日期；
    根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长；
    选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
  6. 一种终端设备，其特征在于，包括：
    用户信息获取单元，用于获取目标用户的用户信息；
    候选协议选取单元，用于将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议；
    推荐系数计算单元，用于基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
    目标协议确定单元，用于基于所述推荐系数，从候选协议中选取目标协议；
    目标协议推送单元，用于将所述目标协议发送给所述目标用户的用户终端。
  7. 根据权利要求6所述的终端设备，其特征在于，终端设备还包括：
    训练数据获取单元，用于获取训练用户的用户信息和历史购买协议的标识；
    学习参数训练单元，用于基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
    Figure PCTCN2018097563-appb-100003
    其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息；I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值；
    候选协议推荐模型生成单元，用于基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
  8. 根据权利要求6所述的终端设备，其特征在于，所述目标协议推送单元包括：
    推送记录节点创建单元，用于以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点；
    协议调整信息导入单元，用于若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
  9. 根据权利要求6所述的终端设备，其特征在于，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述推荐系数计算单元包括：
    平均收入获取单元，用于获取所述目标用户的平均收入金额；
    消费匹配度计算单元，用于根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度；
    回报率计算单元，用于根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；
    归一化单元，用于对所述候选协议的购买数量以及历史用户评分值进行归一化处理；
    参数导入计算单元，用于将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
    Figure PCTCN2018097563-appb-100004
    其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
  10. 根据权利要求6-9任一项所述的终端设备，其特征在于，所述用户信息获取单元包 括：
    有效日期获取单元，用于获取用户数据库中各个用户的已购买协议的有效日期；
    剩余有效时长计算单元，用于根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长；
    目标用户确定单元，用于选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
  11. 一种终端设备，其特征在于，所述终端设备包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机可读指令，所述处理器执行所述计算机可读指令时实现如下步骤：
    获取目标用户的用户信息；
    将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议；
    基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
    基于所述推荐系数，从候选协议中选取目标协议；
    将所述目标协议发送给所述目标用户的用户终端。
  12. 根据权利要求11所述的终端设备，其特征在于，在将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取训练用户的用户信息和历史购买协议的标识；
    基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
    Figure PCTCN2018097563-appb-100005
    其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息；I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值；
    基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
  13. 根据权利要求11所述的终端设备，其特征在于，所述将所述目标协议发送给所述 目标用户的用户终端，包括：
    以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点；
    若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
  14. 根据权利要求11所述的终端设备，其特征在于，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数，包括：
    获取所述目标用户的平均收入金额；
    根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度；
    根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；所述回报金额具体为满足所述候选协议的预设条件时目标用户可获取的金额；
    对所述候选协议的购买数量以及历史用户评分值进行归一化处理；
    将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
    Figure PCTCN2018097563-appb-100006
    其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
  15. 根据权利要求11-14任一项所述的终端设备，其特征在于，所述获取目标用户的用户信息，包括：
    获取用户数据库中各个用户的已购买协议的有效日期；
    根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长；
    选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
  16. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现如下步骤：
    获取目标用户的用户信息；
    将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的 候选协议；
    基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数；
    基于所述推荐系数，从候选协议中选取目标协议；
    将所述目标协议发送给所述目标用户的用户终端。
  17. 根据权利要求16所述的计算机可读存储介质，其特征在于，在将所述用户信息导入候选协议推荐模型，从协议文件库中选取与所述目标用户匹配的候选协议之前，所述处理器执行所述计算机可读指令时还实现如下步骤：
    获取训练用户的用户信息和历史购买协议的标识；
    基于所述用户信息以及所述历史购买协议的标识，调整长短期记忆LSTM神经网络中的学习参数，以使所述学习参数满足以下条件：
    Figure PCTCN2018097563-appb-100007
    其中，θ *为调整后的所述学习参数；S为所述历史购买协议的标识；I为所述用户信息；I 1,I 2,I 3,…,I n为所述用户信息包含的各项用户参数的参数值；n为所述用户参数的个数；p(S|I 1,I 2,I 3,…,I n；θ)为当所述学习参数的值为θ时，将所述训练用户的用户信息导入到所述LSTM神经网络，输出结果为该训练用户的历史购买协议的标识的概率值；max θ∑ (I,S)logp(S|I 1,I 2,I 3,…,I n；θ)为所述概率值取最大值时所述学习参数的取值；
    基于调整学习参数后的所述LSTM神经网络，生成所述候选协议推荐模型。
  18. 根据权利要求16所述的计算机可读存储介质，其特征在于，所述将所述目标协议发送给所述目标用户的用户终端，包括：
    以预设的推送周期向所述用户终端发送所述目标协议，并生成推送记录时间轴；所述推送记录时间轴包含在各个推送时刻所创建的推送记录节点；
    若接收到所述用户终端根据所述目标协议返回的协议调整信息，则把所述协议调整信息导入至所述推送记录节点内；所述协议调整信息用于调整所述目标协议的参数。
  19. 根据权利要求16所述的计算机可读存储介质，其特征在于，所述预设维度包括：购买数量、历史用户评分值、消费匹配度以及回报率；所述基于各个所述候选协议在多个预设维度的评分值，分别计算各个候选协议的推荐系数，包括：
    获取所述目标用户的平均收入金额；
    根据所述平均收入金额以及各个所述候选协议的购买金额，分别计算所述目标用户与各个所述候选协议的消费匹配度；
    根据各个所述候选协议的回报金额以及所述购买金额，分别计算各个所述候选协议的回报率；所述回报金额具体为满足所述候选协议的预设条件时目标用户可获取的金额；
    对所述候选协议的购买数量以及历史用户评分值进行归一化处理；
    将各个所述候选协议归一化后的购买数量、归一化后的历史用户评分值、消费匹配度以及回报率导入推荐系数计算模型，得到各个所述候选协议的推荐系数；所述推荐系数计算模型具体为：
    Figure PCTCN2018097563-appb-100008
    其中，Q为所述推荐系数；SL′为所述归一化后的购买数量；Pnt′为所述归一化后的历史用户评分值；Mth为所述消费匹配度；PB为所述回报率；β为预设的补偿系数。
  20. 根据权利要求16-19任一项所述的计算机可读存储介质，其特征在于，所述获取目标用户的用户信息，包括：
    获取用户数据库中各个用户的已购买协议的有效日期；
    根据所述有效日期以及当前日期，分别计算各个所述已购买协议的剩余有效时长；
    选取所述剩余有效时长小于预设的时长阈值的已购买协议对应的用户作为所述目标用户。
PCT/CN2018/097563 2018-04-08 2018-07-27 一种协议文件的推送方法及终端设备 Ceased WO2019196261A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810305550.2A CN108710634B (zh) 2018-04-08 2018-04-08 一种协议文件的推送方法及终端设备
CN201810305550.2 2018-04-08

Publications (1)

Publication Number Publication Date
WO2019196261A1 true WO2019196261A1 (zh) 2019-10-17

Family

ID=63866536

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/097563 Ceased WO2019196261A1 (zh) 2018-04-08 2018-07-27 一种协议文件的推送方法及终端设备

Country Status (2)

Country Link
CN (1) CN108710634B (zh)
WO (1) WO2019196261A1 (zh)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111461667A (zh) * 2020-04-08 2020-07-28 开封博士创新技术转移有限公司 一种海量数据导入方法、装置、服务器及可读存储介质
CN112187945A (zh) * 2020-09-30 2021-01-05 北京有竹居网络技术有限公司 信息推送方法、装置和电子设备
CN112686646A (zh) * 2021-01-31 2021-04-20 重庆渝高科技产业(集团)股份有限公司 一种合同线上填报管理方法及系统
CN113077352A (zh) * 2021-04-22 2021-07-06 北京十一贝科技有限公司 基于用户信息和保险相关信息的保险服务物品推荐方法
CN113627900A (zh) * 2021-08-10 2021-11-09 未鲲(上海)科技服务有限公司 模型训练方法、装置与存储介质
CN113902481A (zh) * 2021-10-14 2022-01-07 平安银行股份有限公司 权益确定方法、装置、存储介质及设备
CN114338931A (zh) * 2021-12-30 2022-04-12 中国电信股份有限公司 批量短信提醒方法、装置以及计算机可读存储介质
CN115705382A (zh) * 2021-08-12 2023-02-17 腾讯科技（深圳）有限公司 信息推送方法及其装置、存储介质
CN115705577A (zh) * 2021-08-04 2023-02-17 腾讯科技（深圳）有限公司 推荐信息的成本控制方法、装置、计算机设备和存储介质
CN115914408A (zh) * 2022-11-07 2023-04-04 阿里健康科技(中国)有限公司 短链接信息的数据分析方法和管理方法、电子设备
CN116437159A (zh) * 2023-03-14 2023-07-14 深圳感臻智能股份有限公司 一种基于数字电视协议的数据处理方法、系统及介质

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110119877B (zh) * 2019-04-04 2022-12-09 平安科技（深圳）有限公司 一种目标员工的选取方法及设备
CN110704726B (zh) * 2019-08-19 2023-07-28 中国平安财产保险股份有限公司 基于神经网络的数据推送方法、及其相关设备
CN111062709B (zh) * 2019-12-30 2021-11-23 北京三快在线科技有限公司 资源转移方式推荐方法、装置、电子设备及存储介质
CN113656686B (zh) * 2021-07-26 2024-09-06 深圳市中元产教融合科技有限公司 一种基于产教融合的任务报告的生成方法及服务系统
KR102613591B1 (ko) 2022-10-19 2023-12-14 한국전자기술연구원 워크로드별 최적 동기화 브로커 선정 방법 및 시스템

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100030582A1 (en) * 2008-07-29 2010-02-04 Rippel Steven P Centrally maintained portable driving score
CN103888543A (zh) * 2014-04-04 2014-06-25 河南理工大学 基于Web服务的医疗资源推荐方法及系统
CN106204202A (zh) * 2016-06-29 2016-12-07 百度在线网络技术（北京）有限公司 一种车险信息推荐方法和装置
CN107730389A (zh) * 2017-09-30 2018-02-23 平安科技（深圳）有限公司 电子装置、保险产品推荐方法及计算机可读存储介质
US20180082030A1 (en) * 2016-09-19 2018-03-22 International Business Machines Corporation Automatic Adjustment of Treatment Recommendations Based on Economic Status of Patients

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105869024A (zh) * 2016-04-20 2016-08-17 北京小米移动软件有限公司 商品的推荐方法及装置
EP3566182A1 (en) * 2017-02-06 2019-11-13 Deepmind Technologies Limited Memory augmented generative temporal models
CN107506414B (zh) * 2017-08-11 2020-01-07 武汉大学 一种基于长短期记忆网络的代码推荐方法
CN107798579B (zh) * 2017-10-19 2021-06-25 中国平安财产保险股份有限公司 一种协议文件的生成方法及其终端

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100030582A1 (en) * 2008-07-29 2010-02-04 Rippel Steven P Centrally maintained portable driving score
CN103888543A (zh) * 2014-04-04 2014-06-25 河南理工大学 基于Web服务的医疗资源推荐方法及系统
CN106204202A (zh) * 2016-06-29 2016-12-07 百度在线网络技术（北京）有限公司 一种车险信息推荐方法和装置
US20180082030A1 (en) * 2016-09-19 2018-03-22 International Business Machines Corporation Automatic Adjustment of Treatment Recommendations Based on Economic Status of Patients
CN107730389A (zh) * 2017-09-30 2018-02-23 平安科技（深圳）有限公司 电子装置、保险产品推荐方法及计算机可读存储介质

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111461667A (zh) * 2020-04-08 2020-07-28 开封博士创新技术转移有限公司 一种海量数据导入方法、装置、服务器及可读存储介质
CN111461667B (zh) * 2020-04-08 2023-08-18 开封博士创新技术转移有限公司 一种海量数据导入方法、装置、服务器及可读存储介质
CN112187945A (zh) * 2020-09-30 2021-01-05 北京有竹居网络技术有限公司 信息推送方法、装置和电子设备
CN112686646B (zh) * 2021-01-31 2023-09-22 重庆渝高科技产业(集团)股份有限公司 一种合同线上填报管理方法及系统
CN112686646A (zh) * 2021-01-31 2021-04-20 重庆渝高科技产业(集团)股份有限公司 一种合同线上填报管理方法及系统
CN113077352A (zh) * 2021-04-22 2021-07-06 北京十一贝科技有限公司 基于用户信息和保险相关信息的保险服务物品推荐方法
CN113077352B (zh) * 2021-04-22 2024-02-02 北京十一贝科技有限公司 基于用户信息和保险相关信息的保险服务物品推荐方法
CN115705577A (zh) * 2021-08-04 2023-02-17 腾讯科技（深圳）有限公司 推荐信息的成本控制方法、装置、计算机设备和存储介质
CN115705577B (zh) * 2021-08-04 2026-02-27 腾讯科技（深圳）有限公司 推荐信息的推送方法、装置、计算机设备和存储介质
CN113627900A (zh) * 2021-08-10 2021-11-09 未鲲(上海)科技服务有限公司 模型训练方法、装置与存储介质
CN115705382A (zh) * 2021-08-12 2023-02-17 腾讯科技（深圳）有限公司 信息推送方法及其装置、存储介质
CN113902481A (zh) * 2021-10-14 2022-01-07 平安银行股份有限公司 权益确定方法、装置、存储介质及设备
CN114338931A (zh) * 2021-12-30 2022-04-12 中国电信股份有限公司 批量短信提醒方法、装置以及计算机可读存储介质
CN115914408A (zh) * 2022-11-07 2023-04-04 阿里健康科技(中国)有限公司 短链接信息的数据分析方法和管理方法、电子设备
CN116437159A (zh) * 2023-03-14 2023-07-14 深圳感臻智能股份有限公司 一种基于数字电视协议的数据处理方法、系统及介质

Also Published As

Publication number Publication date
CN108710634B (zh) 2023-04-18
CN108710634A (zh) 2018-10-26

Similar Documents

Publication Publication Date Title
WO2019196261A1 (zh) 一种协议文件的推送方法及终端设备
US20220398635A1 (en) Holistic analysis of customer sentiment regarding a software feature and corresponding shipment determinations
CN109242612B (zh) 一种产品推荐的方法及设备
CN102542012B (zh) 分类搜索查询的结果
CN106372249B (zh) 一种点击率预估方法、装置及电子设备
US20240202769A1 (en) Method, apparatus, device, storage medium and program product for promoter determination
US20140324729A1 (en) Computerized System and Method for Determining an Action Person's Influence on a Transaction
CN113935322B (zh) 基于自然语言处理的案件分配方法、装置、设备及介质
CN118013120B (zh) 基于聚类标签优化推荐给用户的产品的方法、介质和设备
US20180293312A1 (en) Computerized Method and System for Organizing Video Files
CN110633410A (zh) 信息处理方法及装置、存储介质、电子装置
WO2019192130A1 (zh) 客户分类方法、装置及存储介质
CN110472143A (zh) 一种信息推送方法、装置、可读存储介质及终端设备
CN109800363A (zh) 构建资信关系网络的方法、装置、设备及存储介质
CN107590742A (zh) 一种基于行为的社交网络用户属性值反演方法
JP2018116694A (ja) 算出装置、算出方法及び算出プログラム
CN118195738A (zh) 数据处理方法、装置及设备
US20220036386A1 (en) Subscription renewal prediction with a cooperative component
US11973832B2 (en) Resolving polarity of hosted data streams
US11379929B2 (en) Advice engine
US20220318327A1 (en) Ranking similar users based on values and personal journeys
TW201935369A (zh) 意見領袖之網路交易管理方法、系統以及儲存媒體
CN108460611A (zh) 一种信息处理方法、及其应用方法和相关设备
CN115391666A (zh) 医院在线推荐方法、装置、计算机设备和存储介质
CN111783445A (zh) 数据生成方法、装置、介质及电子设备

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 02/02/2021)

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18914881

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 18914881

Country of ref document: EP

Kind code of ref document: A1