CN115587170A - Personalized problem item recommendation method, system and storage medium - Google Patents
Personalized problem item recommendation method, system and storage medium Download PDFInfo
- Publication number
- CN115587170A CN115587170A CN202211176715.3A CN202211176715A CN115587170A CN 115587170 A CN115587170 A CN 115587170A CN 202211176715 A CN202211176715 A CN 202211176715A CN 115587170 A CN115587170 A CN 115587170A
- Authority
- CN
- China
- Prior art keywords
- recommendation
- question
- salesman
- value
- algorithm
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06Q30/01—Customer relationship services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
Abstract
The invention relates to a problem item personalized recommendation method, a system and a storage medium, wherein the method comprises the following steps: acquiring a knowledge base, acquiring online behavior information and historical intelligent interaction records of an operator, and adding a label to the operator based on the online behavior information; the method comprises the steps that a plurality of problem recommendation algorithms are applied to carry out problem recommendation for an operator to obtain a plurality of candidate problem sets, the problem recommendation algorithms are used for recommending a preset number of problems for the operator from a knowledge base, and one candidate problem set corresponds to the preset number of problems obtained by one problem recommendation algorithm; fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman; the problem recommendation algorithm comprises a label recall algorithm, a collaborative filtering recall algorithm, a link recall algorithm and a popularity recall algorithm. Compared with the prior art, the method and the system have the advantages that the diversity of the problems to be recommended is increased in a multi-way recall result fusion mode, and the matching degree of the personalized recommendation problems and the knowledge requirements of the service staff is improved.
Description
Technical Field
The invention relates to an intelligent question-answering system, in particular to a question item personalized recommendation method and system for a QA question-answering system of an insurance salesman and a storage medium.
Background
In the current intelligent wave, finance and science and technology become the important strategic direction for the intelligent transformation and upgrading of the traditional financial industry, and based on the characteristics of high specialty, wide related range, fast product updating and the like of insurance products, the intelligent question-answering system also becomes one of the most urgent requirements for optimizing service flow and improving service efficiency transformation of the insurance industry, and is also one of the most direct landing applications of intelligent transformation.
Under the background of an era that the information amount is explosively increased at present, the method can record the online behaviors of the users and mine the intention and preference of the users, can provide powerful support for enterprise operation and marketing, generates a large amount of high-quality and high-value data in the business form of the insurance industry, has great mining and using values, and can recommend the problems meeting the corresponding requirements of the users for different users based on the online behavior track of the users and an intelligent question-answering system. At present, many insurance companies analyze user online behavior data by trying to use models such as collaborative filtering and association rules, and recommend products meeting personal requirements and product-related problem consultation for the insurance companies through an intelligent maintenance mode, so that consultation efficiency and customer experience of customers are greatly improved.
However, in the insurance industry, the main application object of the existing personalized problem recommendation technology is an insurance client, the problem consultation service for recommending corresponding product demand for the insurance client is mainly mined according to the behavior and preference of the insurance client, the current market mainstream scheme is to use the problem with high question frequency within a period of time to display on a first page when the relevant problem is pushed to the salesman, the problem recommendation of thousands of people is not considered according to the knowledge demand of different salesmen, the different knowledge demands of different salesmen are ignored, and the irreplaceable function of the salesmen in the sales track of the insurance industry is ignored.
Firstly, different salesmen have different knowledge requirements, for example, salesmen in different sales channels sell different products, salesmen in different job levels care different exhibition processes, and the like, if undifferentiated problem pushing is performed, the profit is very low, and the problem solving efficiency of the salesmen cannot be improved, and the click rate of the intelligent question-answering platform cannot be improved.
Secondly, the first-line salesman is the most direct contact between a company and a client, and is an insignia of the closest distance between a client appeal and a service pain point, and the direct understanding of a large number of clients on insurance company products mainly depends on the self product knowledge accumulation of an outworker and professional knowledge storage and service capacity of salesman sales skills, so that the understanding degree of the outworker on the insurance products is also related to the marketing performance information of the company.
In addition, in the application of the intelligent question-answering system, intelligent digital marketing is expected to be performed in a way of completely replacing human beings through intelligent care in more application scenes, but many problems are faced in the current application. Firstly, the customer obtaining in the intelligent maintenance mode is passive, the customer logs in a related platform for consultation only when the customer has related product consultation requirements, and a large amount of operation cost is consumed in a new customer drainage link. Secondly, the service scene of intelligent maintenance is limited, and the full life cycle of the insurance user is not covered. The services provided by the mainstream intelligent care products in the market at present are basically developed around how the client selects proper insurance, and belong to pre-sale services. And consultation, payment, renewal, claim settlement and the like involved in sale and after sale are less involved, and the maximum value for maintaining the life cycle of the user is not mined. Thus, the role of the clerk in insurance sales cannot be completely replaced.
In the face of diverse insurance products and complex product terms, insurance traders are faced with a large number of insurance professional business questions from clients each day. With the gradual upgrade of the client insurance requirements, the insurance company operation shows the trend of product diversification and the intellectualization of the process of the business staff, and higher requirements are provided for the professional knowledge and the service capability of the front-line business staff. Therefore, how to cultivate a professional team with more specialization and more complete knowledge reserve is a great important breakthrough for the insurance company to improve the income, so that the method has great business value for providing personalized insurance-related problem consultation for different insurance professionals.
Disclosure of Invention
The technical problem to be solved by the embodiments of the present invention is to provide a problem item personalized recommendation method, system and storage medium, which are used for recommending a problem for an insurance salesperson, increasing the diversity of problems to be recommended by means of multi-path recall result fusion, and improving the matching degree between a personalized recommendation problem and knowledge requirements of the salesperson.
The purpose of the invention can be realized by the following technical scheme:
in order to solve the above problem, an embodiment of the present invention provides a method for recommending a problem item in a personalized manner, including the following steps:
acquiring a knowledge base, wherein the knowledge base comprises a plurality of problems, acquiring online behavior information and historical intelligent interaction records of an operator, and adding a label to the operator based on the online behavior information;
the method comprises the steps that a plurality of problem recommendation algorithms are applied to carry out problem recommendation on a salesperson to obtain a plurality of candidate problem sets, the problem recommendation algorithms are used for recommending a preset number of problems to the salesperson from a knowledge base, and one candidate problem set corresponds to the preset number of problems obtained by one problem recommendation algorithm;
fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman;
the problem recommendation algorithm comprises a label recall algorithm, a collaborative filtering recall algorithm, a link recall algorithm and a popularity recall algorithm.
Further, the problem recommendation algorithm includes a tag recall algorithm, which specifically includes the following:
acquiring a salesman set, acquiring basic attribute information, tags and historical intelligent interaction records of each salesman in the salesman set, and classifying the problems in a knowledge base;
extracting a plurality of attribute variables from each self-basic attribute information and each label, and performing box separation processing on the service member set based on the value number of the attribute variables to obtain a plurality of boxes, wherein each box corresponds to one value of one attribute variable;
calculating the WOE value of each sub-box and each type of problem to obtain each value of each attribute variable and the WOE value of each type of problem, calculating the IV value of each attribute variable and each type of problem, and determining the association degree of each type of problem and each attribute variable;
and recommending a preset number of questions for each salesman in the self-knowledge base according to the value of the attribute variable of the salesman.
Further, the "bin dividing the salesman based on the attribute variables" specifically includes:
traversing all the attribute variables, judging the value of each attribute variable, and determining the value quantity of each attribute variable; carrying out box separation processing on the service member set according to each value of each attribute variable to obtain
Each sub-box, where M is the number of attribute variables, s m The value number of the mth attribute variable is obtained.
Further, in the "calculating the WOE value of each bin and each type of problem", the calculation formula of the WOE value of the ith bin and the jth type of problem is as follows:
wherein, y i Means the number of operators in the ith sub-box who have asked the jth question, y T Means the number of the operators who ask the jth question in the operator set, n i Number of operators in ith sub-box who do not ask jth question, n T Refers to the number of operators in the set who do not ask the jth question.
Further, in the "calculating the IV value of each attribute variable and each type of problem", the calculation formula of the mth attribute variable and the IV value of the jth problem is as follows:
IV ij ＝(Py ij -Pn ij )*WOE ij
wherein s is m Is the value number of the m-th attribute variable, IV ij Is the IV value, IV, of the ith bin and the jth question m,j Is the IV value of the mth attribute variable and the jth problem.
Further, the problem recommendation algorithm includes a collaborative filtering recall algorithm, which is specifically as follows:
acquiring a salesman set, acquiring historical intelligent interaction records of each salesman in the salesman set, and constructing a salesman-question scoring matrix, wherein each row of the salesman-question scoring matrix represents one salesman, each column of the salesman-question scoring matrix represents one question, and the value of each element represents the number of questioning times;
calculating the similarity between the problems by using the Jaccard distance, constructing a similarity matrix, wherein the rows and the columns in the similarity matrix represent the problems, and the value of an element represents the similarity of the two problems;
calculating the recommendation score recommended to each salesman by each problem, and constructing a salesman-problem recommendation matrix:
R U-I ＝M U-I *S I-I
wherein M is U-I Scoring a matrix for the operator-question, S I-I Is a similarity matrix, R U-I Recommending a matrix for the operator-question;
and recommending a preset number of questions for each operator according to the operator-question recommendation matrix.
Further, the similarity calculation formula of the problem a and the problem b is as follows:
where a is the number of times question a is asked and B is the number of times question a is asked.
Further, the step of performing fusion sorting on the problems in the plurality of candidate problem sets, screening out the plurality of problems and recommending the plurality of problems to the operator specifically comprises the following steps:
acquiring the problems in each candidate problem set, and performing duplicate removal on the same problems;
the problems are sorted according to the recommendation sorting under each problem recommendation algorithm, and when the recommendation sorting of the two problems is the same, the two problems are sorted according to the preset algorithm priority;
and selecting and outputting Q questions according to the sorted questions, wherein Q is a preset recommended number.
Another embodiment of the present invention further provides a system for personalized recommendation of question items, including:
a knowledge base containing a plurality of questions;
the data acquisition unit is used for acquiring online behavior information and historical intelligent interaction records of the salesman and adding a label to the salesman based on the online behavior information;
the recommendation unit is used for obtaining a plurality of candidate problem sets and comprises a plurality of sub-recommendation modules, each sub-recommendation module applies a problem recommendation algorithm to recommend problems of a preset number to an operator from a knowledge base, and one candidate problem set corresponds to the problems of the preset number obtained by one sub-recommendation module;
and the output unit is used for fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman.
Yet another embodiment of the present invention further provides a storage medium, which has a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the method for personalized recommendation of question items according to the first aspect of the present invention.
Compared with the prior art, the invention has the following beneficial effects:
(1) The invention deeply excavates the online behavior characteristics of insurance operators, simultaneously carries out personalized problem recommendation for the insurance operators by using a multi-channel recall strategy by combining relevant personal attributes, fuses and sequences recommendation results of all channels to obtain personalized problem push matched with knowledge storage and knowledge requirements of different operators, creates a special knowledge service platform for the operators, and improves the matching degree of the personalized recommendation problem and the knowledge requirements of the operators.
(2) A label recalling algorithm is designed, the WOE value and the IV value are applied to a problem recommendation scene to judge the association degree of the attribute variables and the problems, and the association degree of the attribute variables and the problems is judged according to the association degree, so that the attribute characteristics highly matched with the problem labels are selected, and the related problems matched with the knowledge requirements of the personnel are predicted more accurately from the personal attribute characteristics of the salesman.
(3) Audience crowds of different recall schemes have differences, and accordingly collaborative filtering recall, link recall, label recall and popularity recall are performed respectively by utilizing interactive data with different durations, so that audience areas of crowds to be recommended are increased, and diversity of problems to be recommended is improved.
Drawings
FIG. 1 is a flow chart of a method for personalized recommendation of question items;
fig. 2 is a model framework of a problem item personalized recommendation system.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
The present specification provides method steps as in the examples or flow diagrams, but may include more or fewer steps based on routine or non-inventive labor. The order of steps recited in the embodiments is merely one manner of performing the steps in a multitude of orders and does not represent the only order of execution. In actual system or server product execution, the method shown in the embodiment or the figures can be executed sequentially or in parallel (for example, in the environment of parallel processors or multi-thread processing), or the execution sequence of steps without timing limitation can be adjusted.
Example 1:
a method for personalized problem item recommendation for insurance officers, hereinafter also referred to as officers, agents or users, as shown in fig. 1, comprising the steps of:
s1, acquiring a knowledge base, wherein the knowledge base comprises a plurality of problems, acquiring online behavior information and historical intelligent interaction records of an operator, and adding a label to the operator based on the online behavior information;
wherein the knowledge base is the knowledge base of the QA question-answering system. Historical intelligent interaction records refer to historical interaction records of an operator using a QA question-and-answer system. The online behavior information of the salesman comprises online behavior track information such as APP frequency, visitor micro frequency, order-giving transaction system use frequency and the like, effective features such as historical intelligent interaction use times, historical worksheet use times, historical groove spitting use times, historical outer chain library click times and the like are extracted from the online behavior track information, the salesman is grouped, and tags are added.
S2, recommending questions for the service personnel by applying multiple question recommendation algorithms to obtain multiple candidate question sets, wherein the question recommendation algorithms are used for recommending a preset number of questions to the service personnel from a knowledge base, and one candidate question set corresponds to the preset number of questions obtained by one question recommendation algorithm;
in the embodiment, 4 problem recommendation algorithms including a tag recall algorithm, a collaborative filtering recall algorithm, a link recall algorithm and a popularity recall algorithm are designed, and in other embodiments, the types and the number of the problem recommendation algorithms can be adjusted according to needs.
And obtaining a first candidate problem set by using a collaborative filtering recall algorithm, obtaining a second candidate problem set by using a link recall algorithm, obtaining a third candidate problem set by using a label recall algorithm, and obtaining a fourth candidate problem set by using a popularity recall algorithm. And finally, fusing and sequencing the candidate problems in all the candidate problem sets to find out the final recommendation problem.
(1) The tag recall algorithm is as follows:
(1) acquiring a salesman set, acquiring basic attribute information, tags and historical intelligent interaction records of each salesman in the salesman set, and classifying the problems in a knowledge base;
the basic attribute information of the salesman includes gender, academic history, marital status, current month FYC, working age, business background, business level, star rating, and the like, and the questions in the knowledge base can be divided into a plurality of categories, in this embodiment, the questions in the knowledge base are divided into five categories: operational knowledge, product knowledge, activity knowledge, practice knowledge, and other knowledge.
(2) Extracting a plurality of attribute variables from each self-basic attribute information and each label, and performing box separation processing on the service member set based on the value number of the attribute variables to obtain a plurality of boxes, wherein each box corresponds to one value of one attribute variable;
extracting attribute variables according to the basic attribute information, the tags and the historical intelligent interaction records, wherein in the embodiment, 15 attribute variables are extracted, and the method comprises the following steps: sales channel, state of salesman, whether healthy manpower is available, whether performance manpower is available, whether CA manpower is available, whether CE manpower is available, whether 1/4MDRT mark is available, whether 1/2MDRT mark is available, whether 3/4MDRT mark is available, whether two thousand P manpower is available, whether performance manpower is available, whether supervisor is available, whether reach-up manpower is available, region type and gender.
The box separation treatment specifically comprises the following steps:
traversing all the attribute variables, judging the value of each attribute variable, and determining the value number of each attribute variable, wherein if the gender is male or female, the value number is 2, the age can be divided into age groups, such as under 30 years, 30-40 years, 41-50 years and the like, and the value number is the number of the divided age groups. Carrying out box separation processing on the service member set according to each value of each attribute variable to obtain
Each sub-box, wherein M is the number of attribute variables, s m The value number of the mth attribute variable is obtained.
(3) Calculating the WOE value of each sub-box and each type of problem to obtain each value of each attribute variable and the WOE value of each type of problem, calculating the IV value of each attribute variable and each type of problem, and determining the association degree of each type of problem and each attribute variable;
WOE (Weight of Evidence), the WOE value of the ith bin and the jth question, is calculated as follows:
wherein, y i The number of the operators who ask the jth question in the ith sub-box, y T Means the number of the operators who ask the jth question in the operator set, n i Number of operators in ith sub-box who do not ask jth question, n T Refers to the number of operators in the set who do not ask the jth question.
The IV (information value) is used to measure the prediction capability of the attribute variables, and the calculation formula of the IV value of the mth attribute variable and the jth question is as follows:
IV ij ＝(Py ij -Pn ij )*WOE ij
wherein s is m Is the value number, IV, of the mth attribute variable ij IV value for ith bin and jth problem, IV m,j Is the IV value of the mth attribute variable and the jth problem.
The operator who asks the jth question is called the responding client, in fact, py ij Is the proportion of responding clients in the sub-box to the clients in the centralized response of the business staff, pn ij Is the proportion of non-responding customers in the sub-box to non-responding customers in the crew set. The WOE value is actually the difference between the proportion of the responding client to all responding clients in the current bin and the proportion of the non-responding client to all non-responding clients in the current bin, and the higher the absolute value of the WOE value is, the larger the difference between the proportions of the responding client and the non-responding client in the bin is, namely, the greater the influence of the bin on the classification results of the responding client and the non-responding client is。
The larger the value of IV, the higher the predictive ability of the attribute variable for the question whether to ask the classification.
Taking the operation knowledge category problem as an example, the IV value between the operator state attribute variable and the operation knowledge category problem is high, that is, the correlation degree between the operator state attribute variable and the operation knowledge category problem is high, so when the recommendation of the operation knowledge category problem is considered, the operator state attribute variable can be used as a basis, other attribute variables do not need to be considered, the number of question records of the first-class classification problem corresponding to each attribute label enumeration value of the outside operator in 30 days in history is counted, the problem before the question rank of 20 is screened under each enumeration value, and the corresponding problem is recommended according to the operator state attribute characteristic corresponding to the operator.
(4) And recommending a preset number of questions for each salesman in the self-knowledge base according to the value of the attribute variable of the salesman.
Through the calculation of the WOE value and the IV value, the attribute variable with higher corresponding contribution degree can be found under each problem classification, therefore, a plurality of problems corresponding to the values taken by the attribute variables of the amateur are merged, the problems are sorted according to the occurrence times of the problem history, and the TOP20 problem finally sorted is taken as a candidate problem set recalled by the label.
(2) The Item-Based collaborative filtering recall algorithm recommends a new problem (not containing the user history interest problem) interested in the recommended user according to the similarity between the standard problems and the user history interest problem, and the details are as follows:
acquiring a salesman set, acquiring historical intelligent interaction records of each salesman in the salesman set, and constructing a salesman-question scoring matrix, wherein each row of the salesman-question scoring matrix represents one salesman, each column of the salesman-question scoring matrix represents one question, and the value of an element represents the number of questioning times;
calculating the similarity between the problems by using the Jaccard distance, constructing a similarity matrix, wherein the rows and the columns in the similarity matrix represent the problems, the value of an element represents the similarity of the two problems, and the similarity calculation formula of the problem a and the problem b is as follows:
wherein, A is the number of times of question a being asked, B is the number of times of question a being asked;
calculating the recommendation value recommended to each operator by each problem, and constructing an operator-problem recommendation matrix, wherein the rows in the operator-problem recommendation matrix represent users, and the columns represent problems:
R U-I ＝M U-I *S I-I
wherein M is U-I Scoring a matrix for the operator-question, S I-I Is a similarity matrix, R U-I Recommending a matrix for the operator-question;
according to the operator-question recommendation matrix, a preset number of questions are recommended for each operator, in this embodiment, a row of elements of the operator-question recommendation matrix may be sorted, and a TOP20 standard question is selected as a recommended question of the operator corresponding to the row of elements.
(3) Link recall algorithm and heat recall algorithm:
the intelligent interactive questioning historical records of the service personnel may have a certain degree of relevance, and by using a link recalling method and analyzing the relevance between the first-degree question and the second-degree question of the user, the online precursor and subsequent behavior relation links of the service personnel can be further mined. Through analyzing the intelligent interaction records of the first-degree question and the second-degree question combined ranking of the salesperson in two months of history, and combining the historical question records of the salesperson, relevant question recommendation is carried out on the salesperson, and each TOP20 standard question sentence of each person is obtained.
In the popularity recall algorithm, the problem question history records among users can be found to have certain distribution rules on time and channels through the analysis of intelligent interaction data, and corresponding hotspot recommendation can be performed on the users through the summary of hotspot problems of the time and the channels. And counting the intelligent interaction records of one month of history of all the outworkers, and performing question-asking popularity ranking on all the questions to obtain a TOP20 popularity question.
S3, fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman, wherein the method specifically comprises the following steps:
acquiring the problems in each candidate problem set, and removing the duplicate of the same problem; the problems are sorted according to the recommendation sorting under each problem recommendation algorithm, and when the recommendation sorting of the two problems is the same, the two problems are sorted according to the preset algorithm priority; and selecting and outputting Q questions according to the sorted questions, wherein Q is the preset recommended quantity.
Specifically, 20 problems of each path of each salesman obtained in the collaborative filtering recall, the link recall and the label recall are fused and sequenced, and the fusion rule mainly comprises the following steps:
(1) problem fusion (recall only TOP20 problems per round);
(2) duplicate removal of the same problem in each recall (keeping a record of the source of the problem with higher rank);
(3) ranking the recall problems of each path (preferably, recommending and ranking according to the ranking of the problems in each recall path, and ranking according to the priority of ' collaborative filtering ' > ' link recall ' > ' label recall ' > ' popularity recall ' > ' when the ranking of the problems is the same);
(4) selecting a recommendation problem (selecting a problem with a ranking TOP20 in a recommendation result for recommendation presentation);
through the use condition analysis of the intelligent question-answering platform, about 80% of the field staff who use the interactive function on the intelligent question-answering platform daily currently have historical interactive use records in the last half year, about 20% of the field staff are new staff who have no intelligent interactive use records or use intelligent interaction for the first time in the last half year, wherein the audience population of different recall schemes has differences:
in order to better improve the problem solving capability of a salesman and the problem solving efficiency, the online behavior characteristics of an insurance salesman are deeply excavated, meanwhile, a multi-path recall strategy is used for carrying out personalized problem recommendation for the salesman in combination with relevant personal attributes, the recommendation results of all paths are fused and sequenced to obtain personalized problem push matched with the knowledge storage knowledge requirements of different salesmen, a special knowledge service platform is created for the salesman, and the online customer operation activities are enabled by means of intelligent technical means.
The invention changes the application object in the prior art from an insurance client to an insurance salesman, performs knowledge requirement analysis by combining online behavior data and personal attribute information of the salesman, increases the diversity of the problems to be recommended by a multi-path recall result fusion mode, and improves the matching degree of the personalized recommendation problem and the knowledge requirement of the salesman, and the specific technical scheme has the advantages that:
1. the online behavior and personal attribute characteristics of the service staff are fully utilized:
valuable information hidden in the on-line behavior information of the salesman is fully mined, problem preference of the salesman and product preference of a corresponding client are mined, the salesman is grouped according to the on-line behavior information of the salesman, a label obtained by grouping is incorporated into a characteristic used for label recalling, and recommendation accuracy is improved.
2. By providing personalized professional problem recommendation for the salesperson, professional knowledge storage of the salesperson is improved, and cost reduction and efficiency improvement are realized for the insurance company.
The maximum value of the personalized recommendation algorithm based on the QA question-answering system used in the invention is to enable outworkers, and according to the recommendation algorithm, personalized problem intelligent sequencing is carried out according to a set rule, so that personalized problems meeting personal requirements of each operator are recommended to each operator, rather than independently solving numerous problems of errors and uncertainty at present by a machine. The QA question-answering system-based personalized question recommendation for the salesperson is based on an intelligent product which can meet the requirements of an enterprise on reducing the labor cost and improving the working efficiency, and enables the salesperson to work for outworkers. The maximum implicit value is data accumulation after standardization is achieved in an actual scene, information value appearing in direct communication between a salesman and a client can be mined by mining the question and answer behavior information of the salesman, cost reduction and efficiency improvement are achieved, and accurate marketing and product upgrading can be served subsequently.
3. The audience area of the people to be recommended is increased and the diversity of the problems to be recommended is improved in a multi-way recall mode:
3.1 collaborative filtering recall: the method is characterized in that the method is to recommend questions similar to historical question questions (including similar user interests) to 80% of outworkers who have historical intelligent interaction use records in half a year;
3.1 Link recall: 60% of the historical old clerks who asked 1 to 2 questions within a half year tend to be recommended more business high referential questions.
3.3 tag recall:
(i) The same or similar standard problems tend to be recommended to similar staff groups, and the staff groups are recommended, so that the method has certain service guiding significance;
(ii) It is applicable to new salesman (about 20%) recommendations that first enter the platform.
3.4 Heat recall:
(i) The diversity of the recommendation problems of the old historical servicers is increased;
(ii) High heat issue recommendations are provided for the first daily entry (about 20%) to the platform.
It should be noted that the present application may be implemented in software and/or a combination of software and hardware, for example, implemented using Application Specific Integrated Circuits (ASICs), general purpose computers or any other similar hardware devices. In one embodiment, the software programs of the present application may be executed by a processor to implement the steps or functions described above. Likewise, the software programs (including associated data structures) of the present application may be stored in a computer readable recording medium, such as RAM memory, magnetic or optical drive or diskette and the like. Additionally, some of the steps or functions of the present application may be implemented in hardware, for example, as circuitry that cooperates with the processor to perform various steps or functions.
In addition, some of the present application may be implemented as a computer program product, such as computer program instructions, which when executed by a computer, may invoke or provide methods and/or techniques in accordance with the present application through the operation of the computer. Program instructions which invoke the methods of the present application may be stored on a fixed or removable recording medium and/or transmitted via a data stream on a broadcast or other signal-bearing medium and/or stored within a working memory of a computer device operating in accordance with the program instructions. An embodiment according to the present application herein comprises an apparatus comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein the computer program instructions, when executed by the processor, trigger the apparatus to perform a method and/or solution according to embodiments of the present application as described above.
Another embodiment of the present invention further provides a system for personalized recommendation of question items, including:
a knowledge base containing a plurality of questions;
the data acquisition unit is used for acquiring online behavior information and historical intelligent interaction records of the salesman and adding labels to the salesman based on the online behavior information;
the recommendation unit is used for obtaining a plurality of candidate problem sets and comprises a plurality of sub-recommendation modules, each sub-recommendation module applies a problem recommendation algorithm to recommend problems of a preset number to an operator from a knowledge base, and one candidate problem set corresponds to the preset number of problems obtained by one sub-recommendation module;
and the output unit is used for fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman.
The system for problem item personalized recommendation can be further divided into a data layer, a model layer and an application layer, as shown in fig. 2, the data layer obtains on-line behavior information, historical intelligent interaction records, basic attribute information and the like of a salesman through a human management system, a service support platform and the like, the model layer constructs a problem personalized recommendation model framework, 4 problem recommendation algorithms including a label recall algorithm, a collaborative filtering recall algorithm, a link recall algorithm and a heat recall algorithm are used for recommendation, then recommendation results are fused, and the application layer mainly carries out on-line evaluation of A/B Test (evaluation indexes include PCTR, UCTR and the like), so that the problem personalized recommendation model framework is updated in an iterative manner, model deployment and production are carried out, service preference is explored, and the use activity of a QA question-answering system is improved. The whole system is built and deployed as follows:
(1) data acquisition: the method comprises the steps of obtaining online behavior information, historical intelligent interaction records and the like of a salesman through an intelligent question answering APP interactive embedded point, and obtaining basic attribute information of the salesman up to the current day, such as: gender, academic calendar, marital status, current month FYC, etc.
(2) Establishing a model: analyzing by using the historical 180-day interactive data of the outworker in combination with a collaborative filtering algorithm, and recalling through the interactive problem association degree; using 30-day historical interaction data of outworkers, combining with the operator attribute labels and the problem popularity ranking, and exploring the operators to recall the knowledge preference; and (3) exploring the continuous interaction behavior habit of the user by using 60-day historical interaction data of the outworker in combination with a link analysis method. And performing fusion sequencing on the recommendation problems of all paths in a multi-path recall mode to obtain a daily to-be-recommended problem result table.
(3) Landing and deployment: when a salesman visits the platform every day, information such as a corresponding job number, internal and external work identifiers and the like is obtained through an interface, a salesman recommendation problem result list is called through interface input parameter matching, and 20 problems to be recommended corresponding to each salesman are obtained and displayed on a foreground interface. And the click condition of the salesman to the current day recommendation problem is also used as second day model iteration data to update the knowledge requirement of the salesman in real time.
Still another embodiment of the present invention provides a storage medium, which has a plurality of instructions adapted to be loaded by a processor to perform the steps of the above-mentioned method for personalized recommendation of question items.
Storage media, including persistent and non-persistent, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase-Change RAM (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read-Only Memory (EEPROM), flash Memory or other Memory technology, compact Disc Read-Only Memory (CD-ROM), digital Versatile Disc (DVD) or other optical storage, magnetic cassette tape, magnetic tape storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory computer readable media (transient media), such as modulated data signals and carrier waves.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations can be devised by those skilled in the art in light of the above teachings. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A problem item personalized recommendation method is characterized by comprising the following steps:
acquiring a knowledge base, wherein the knowledge base comprises a plurality of problems, acquiring online behavior information and historical intelligent interaction records of an operator, and adding a label to the operator based on the online behavior information;
the method comprises the steps that a plurality of problem recommendation algorithms are applied to carry out problem recommendation on a salesperson to obtain a plurality of candidate problem sets, the problem recommendation algorithms are used for recommending a preset number of problems to the salesperson from a knowledge base, and one candidate problem set corresponds to the preset number of problems obtained by one problem recommendation algorithm;
fusing and sequencing the problems in the candidate problem sets, screening out a plurality of problems and recommending the problems to a salesman;
the problem recommendation algorithm comprises a label recall algorithm, a collaborative filtering recall algorithm, a link recall algorithm and a popularity recall algorithm.
2. The method for personalized recommendation of question items according to claim 1, wherein said tag recall algorithm is as follows:
acquiring a salesman set, acquiring basic attribute information, tags and historical intelligent interaction records of each salesman in the salesman set, and classifying the problems in a knowledge base;
extracting a plurality of attribute variables from each self-basic attribute information and each label, and performing box separation processing on the service member set based on the value number of the attribute variables to obtain a plurality of boxes, wherein each box corresponds to one value of one attribute variable;
calculating the WOE value of each sub-box and each type of problem to obtain each value of each attribute variable and the WOE value of each type of problem, calculating the IV value of each attribute variable and each type of problem, and determining the association degree of each type of problem and each attribute variable;
and recommending a preset number of questions for each salesman in the self-knowledge base according to the value of the attribute variable of the salesman.
3. The method for recommending problem item individualization according to claim 2, wherein the step of performing box-dividing processing on the salesman based on the attribute variables specifically comprises:
traversing all the attribute variables, judging the value of each attribute variable, and determining the value quantity of each attribute variable; carrying out box separation processing on the service personnel set according to each value of each attribute variable to obtain
Each sub-box, where M is the number of attribute variables, s m The value number of the mth attribute variable is obtained.
4. The method for personalized recommendation of problem item as claimed in claim 3, wherein in said "calculating WOE value of each bin and each type of problem", the calculation formula of WOE value of ith bin and jth type of problem is as follows:
wherein, y i Means the number of operators in the ith sub-box who have asked the jth question, y T Means the number of the salesmen who have asked the jth question in the salesmen i Means the number of operators in the ith sub-box who have not asked the jth question, n T Refers to the number of operators in the set who do not ask the jth question.
5. The method for personalized recommendation of question items according to claim 4, wherein in said "calculating the IV value of each attribute variable and each type of question", the calculation formula of the mth attribute variable and the IV value of the jth question is as follows:
IV ij ＝(Py ij -Pn ij )*WOE ij
wherein s is m Is the value number of the m-th attribute variable, IV ij IV value for ith bin and jth problem, IV m,j Is the IV value of the mth attribute variable and the jth problem.
6. The method for personalized recommendation of problem items according to claim 1, wherein the collaborative filtering recall algorithm is specifically as follows:
acquiring a salesman set, acquiring historical intelligent interaction records of each salesman in the salesman set, and constructing a salesman-question scoring matrix, wherein each row of the salesman-question scoring matrix represents one salesman, each column of the salesman-question scoring matrix represents one question, and the value of each element represents the number of questioning times;
calculating the similarity between the problems by using the Jaccard distance, constructing a similarity matrix, wherein the rows and the columns in the similarity matrix represent the problems, and the value of an element represents the similarity of the two problems;
calculating the recommendation score recommended to each salesman by each problem, and constructing a salesman-problem recommendation matrix:
R U-I ＝M U-I *S I-I
wherein M is U-I Scoring a matrix for the operator-question, S I-I Is a similarity matrix, R U-I Recommending a matrix for the operator-question;
and recommending a preset number of questions for each operator according to the operator-question recommendation matrix.
7. The method for personalized recommendation of question items according to claim 1, wherein the similarity calculation formula of the question a and the question b is as follows:
wherein, A is the number of times the question a is asked, and B is the number of times the question a is asked.
8. The method for personalized recommendation of problem items according to claim 1, wherein the step of performing fusion sorting on the problems in the plurality of candidate problem sets, screening out the plurality of problems and recommending the problems to the operator specifically comprises the steps of:
acquiring the problems in each candidate problem set, and removing the duplicate of the same problem;
the problems are sorted according to the recommendation sorting under each problem recommendation algorithm, and when the recommendation sorting of the two problems is the same, the problems are sorted according to the preset algorithm priority;
and selecting and outputting Q questions according to the sorted questions, wherein Q is the preset recommended quantity.
9. A system for personalized recommendation of question items, comprising:
a knowledge base containing a plurality of questions;
the data acquisition unit is used for acquiring online behavior information and historical intelligent interaction records of the salesman and adding a label to the salesman based on the online behavior information;
the recommendation unit is used for obtaining a plurality of candidate problem sets and comprises a plurality of sub-recommendation modules, each sub-recommendation module applies a problem recommendation algorithm to recommend problems of a preset number to an operator from a knowledge base, and one candidate problem set corresponds to the problems of the preset number obtained by one sub-recommendation module;
and the output unit is used for fusing and sequencing the problems in the candidate problem sets, screening out the problems and recommending the problems to the salesman.
10. A storage medium having a plurality of instructions adapted to be loaded by a processor to perform the steps of the method for personalized recommendation of problem items according to any of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211176715.3A CN115587170A (en) | 2022-09-26 | 2022-09-26 | Personalized problem item recommendation method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211176715.3A CN115587170A (en) | 2022-09-26 | 2022-09-26 | Personalized problem item recommendation method, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115587170A true CN115587170A (en) | 2023-01-10 |
Family
ID=84778280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211176715.3A Pending CN115587170A (en) | 2022-09-26 | 2022-09-26 | Personalized problem item recommendation method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115587170A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117312678A (en) * | 2023-10-24 | 2023-12-29 | 亿企查科技有限公司 | Intelligent recommendation method, equipment and storage medium for potential clients based on big data |
-
2022
- 2022-09-26 CN CN202211176715.3A patent/CN115587170A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117312678A (en) * | 2023-10-24 | 2023-12-29 | 亿企查科技有限公司 | Intelligent recommendation method, equipment and storage medium for potential clients based on big data |
| CN117312678B (en) * | 2023-10-24 | 2024-04-05 | 亿企查科技有限公司 | Intelligent recommendation method, equipment and storage medium for potential clients based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Saura et al. | Setting B2B digital marketing in artificial intelligence-based CRMs: A review and directions for future research | |
| Abusweilem et al. | The impact of knowledge management process and business intelligence on organizational performance | |
| Olszak et al. | Approach to building and implementing business intelligence systems | |
| Leventhal | An introduction to data mining and other techniques for advanced analytics | |
| Yin et al. | A systematic review on business analytics | |
| McBride | Business intelligence in magazine distribution | |
| Savytska et al. | Applied research of digital readiness of retails | |
| Năstase et al. | A new business dimension-business analytics. | |
| EP3281167A1 (en) | Qualitatively planning, measuring, making effecient and capitalizing on marketing strategy | |
| Putler et al. | Customer and Business Analytics | |
| Hosseini et al. | Identifying multi-channel value co-creator groups in the banking industry | |
| Ahmed et al. | Understanding the business value creation process for business intelligence tools in the UAE | |
| Bąska et al. | Identification of advanced data analysis in marketing: A systematic literature review | |
| Khakpour et al. | Visual analytics for decision support: A supply chain perspective | |
| Kadre et al. | Practical Business Analytics Using SAS: A Hands-on Guide | |
| Rao et al. | BMSP-ML: big mart sales prediction using different machine learning techniques | |
| CN115587170A (en) | Personalized problem item recommendation method, system and storage medium | |
| Faiz et al. | Machine Learning Price Prediction During and Before COVID-19 and Consumer Buying Behavior | |
| CN115619571A (en) | Financing planning method, system and device | |
| Chebli et al. | Measuring activity in big data: new estimates of big data employment in the UK market sector | |
| Fu et al. | The Value of Social Media Data in Fashion Forecasting | |
| Nabavi et al. | Introduction to business analytics | |
| Saraswathi et al. | A survey on data mining trends, applications and techniques | |
| Al-alawi et al. | Customer relationship management: the application of data mining techniques in the telecommunications sector | |
| Almtiri et al. | Impact of business analytics and decision support systems on E-commerce in SMEs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |