CN112330391B

CN112330391B - Product recommendation method based on clients and employees

Info

Publication number: CN112330391B
Application number: CN202011154313.4A
Authority: CN
Inventors: 胡亮
Original assignee: Wuhan Century Technology Co ltd
Current assignee: Wuhan Century Technology Co ltd
Priority date: 2020-10-26
Filing date: 2020-10-26
Publication date: 2022-07-08
Anticipated expiration: 2040-10-26
Also published as: CN112330391A

Abstract

The invention provides a product recommendation method and system based on clients and staff, wherein under the condition that the intention and the label of a client are unknown, the method recommends the product with the highest possibility to the client by setting the label and calculating the possibility of the transaction between the product and the client through historical data, so that the transaction rate and the working efficiency of business staff can be improved even if the recommended product cannot be most suitable for the client, and meanwhile, the client demand and the staff benefit are considered.

Description

Product recommendation method based on clients and employees

Technical Field

The invention relates to the technical field of product recommendation, in particular to a product recommendation method based on clients and employees.

Background

When recommending products to customers, business personnel can only recommend proper products to the customers according to the visual image, age and sex of the customers or directly determine the product types according to the requirements expressed by the customers. The former often can't do to recommend the most suitable product for the customer, the latter also can't recommend the most suitable product for the customer when customer's own demand is unclear. At present, most suitable products can be recommended to customers through big data and a machine learning algorithm, but most of the recommendation methods only start from the perspective of the customers, only the recommended products are most suitable for the customers, when the purchasing habits of the customers cause a long transaction period or business personnel are not familiar with the products, and the like, the transaction rate of certain business personnel cannot be improved even if the recommended products are most suitable for the customers.

Disclosure of Invention

In view of the above, on the one hand, the invention provides a product recommendation method based on clients and employees to solve the problem that the conventional product recommendation method cannot improve the transaction rate of a certain business personnel only from the client perspective.

The technical scheme of the invention is realized as follows: a method for customer and employee based product recommendation, comprising:

step S1, setting a plurality of client labels and a plurality of employee labels, and setting a client score of each client label relative to each product and an employee score of each employee label relative to each product according to historical data;

step S2, obtaining a client portrait of a target client and an employee portrait of a specific employee, the client portrait including a plurality of client tags, the employee portrait including a plurality of employee tags;

step S3, constructing a customer rating matrix of the customer portrait relative to all products and an employee rating matrix of the employee portrait relative to all products;

step S4, calculating the column vectors of the predicted tendency values of the target customer to all products and the column vectors of the true tendency values of all products according to the customer scoring matrix, and calculating the column vectors of the predicted familiarity of the specific staff to all products and the column vectors of the true tendency values of all products according to the staff scoring matrix;

step S5, calculating the recommendation values of all products to target customers according to the column vectors of the predicted tendency values and the column vectors of the true tendency values of all products, and calculating the recommendation values of all products to specific employees according to the column vectors of the predicted familiarity and the column vectors of the true tendency values of all products;

and step S6, multiplying the recommended value of each product to the target customer and the recommended value of the product to the specific employee to obtain the success values of all the products, and recommending the product with the highest success value to the target customer.

Optionally, the customer tag includes the nature of the customer, the industry and the address of the customer.

Optionally, the employee tag includes a gender, an enrollment length, a subscription status, and a survey result.

Optionally, step S4 includes:

the method comprises the steps of obtaining information of a product and average scores of target customers on the product, compiling a get _ item _ info function, obtaining the information of the product from a mysql database, and putting the information into an item _ dist;

compiling a get _ train _ data function to extract a training sample from a customer scoring matrix, and returning a list (customerId, Itemid, label), wherein the customerId is a customer portrait ID, the Itemid is a product ID, and the label is an identifier of a positive sample and a negative sample;

model training is carried out on training samples formed by list (customerId, Itemid, label), and column vectors of predicted tendency values of all products and column vectors of true tendency values of all products of a target client are obtained.

Optionally, the training samples include positive samples and negative samples.

Optionally, the learning rate of model training gradually decreases with the iteration round.

Compared with the prior art, the product recommendation method based on the clients and the employees has the following beneficial effects:

(1) under the condition that the intention of a customer and a demand label for a product are unknown, the product label is set, the possibility of interaction between the product and the customer is calculated through historical data, and the product with the highest possibility is recommended to the customer, so that even if the recommended product cannot be most suitable for the customer, the interaction rate and the working efficiency of business personnel can be improved, and meanwhile, the customer demand and the staff benefit are considered;

(2) a scoring matrix is established, missing values in the scoring matrix are predicted according to historical data, and then recommended values of products are obtained, so that high reliability of the recommended values can be guaranteed;

(3) by calculating the distances between the predicted tendency value and the real tendency value and between the predicted familiarity and the real tendency value and selecting the product with the minimum distance for recommendation, the product can be ensured to be suitable for customers as much as possible on the basis of improving the employee transaction rate, and the customer satisfaction is improved.

On the other hand, the invention also provides a product recommendation system based on the client and the staff, so as to solve the problem that the traditional product recommendation system can not improve the transaction rate of a certain business personnel only from the client perspective.

The technical scheme of the invention is realized as follows: a customer and employee based product recommendation system comprising:

the label and score setting module is used for setting a plurality of client labels and a plurality of employee labels, and setting the client score of each client label relative to each product and the employee score of each employee label relative to each product according to historical data;

the system comprises a portrait acquisition module, a portrait acquisition module and a display module, wherein the portrait acquisition module is used for acquiring a client portrait of a target client and an employee portrait of a specific employee, the client portrait comprises a plurality of client tags, and the employee portrait comprises a plurality of employee tags;

the system comprises a scoring matrix construction module, a scoring matrix construction module and a scoring matrix generation module, wherein the scoring matrix construction module is used for constructing a customer scoring matrix of a customer portrait relative to all products and an employee scoring matrix of an employee portrait relative to all products;

the column vector calculation module is used for calculating the column vectors of the predicted tendency values of the target customer to all products and the column vectors of the real tendency values of all products according to the customer scoring matrix, and calculating the column vectors of the predicted familiarity of the specific staff to all products and the column vectors of the real tendency values of all products according to the staff scoring matrix;

the recommendation value calculation module is used for calculating recommendation values of all products for target customers according to the column vectors of the predicted tendency values and the column vectors of the true tendency values of all the products, and calculating recommendation values of all the products for specific employees according to the column vectors of the predicted familiarity and the column vectors of the true tendency values of all the products;

and the success value calculation module is used for multiplying the recommendation value of each product to the target customer and the recommendation value of the product to the specific employee to obtain the success values of all the products and recommending the product with the highest success value to the target customer.

Compared with the prior art, the product recommendation system based on the clients and the employees has the same advantages as the product recommendation method based on the clients and the employees, and the detailed description is omitted.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of a customer and employee based product recommendation method of the present invention;

FIG. 2 is a table of customer ratings for each customer label versus each product of the present invention;

FIG. 3 is a table of employee ratings for each employee tag relative to each product of the present invention;

FIG. 4 is a customer rating matrix of the present invention;

FIG. 5 is an employee scoring matrix of the present invention;

FIG. 6 is a block diagram of a client and employee based product recommendation system of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.

As shown in fig. 1, the product recommendation method based on clients and employees of the present embodiment includes:

As shown in fig. 2, the client tag includes the nature of the client, the industry to which the client belongs, the address where the client is located, and the like, and as shown in fig. 3, the employee tag includes the sex, the duration of the job, the contract status, the investigation result, and the like.

The predicted tendency value represents the likeness or tendency degree of a customer to a certain product label (such as atmosphere and cost performance), the predicted familiarity represents the familiarity of an employee to a certain product, and the real tendency value represents the conformity degree of the product to the certain label (such as atmosphere and cost performance).

Generally, when a salesperson faces different organizations, especially new organizations, the initial 1 st and 2 nd communication is particularly important, and the salesperson needs to quickly find the products interested in the organizations for accurate marketing. The salespersons of a company also have personal characteristics of each person, such as the length of time spent, the degree of understanding of the product, the contract situation, and the like. Suppose that the customer's demand for the product is "atmosphere". "cost performance", if the predicted tendency value of the product a is 0.8, the true tendency value corresponding to the predicted tendency value is 0.7, the predicted familiarity is 0.9, and the true tendency value corresponding to the predicted familiarity is 0.5, i can easily find that the recommended value R of the product a to the customer is 0.8x0.7+0.9x0.5 or 1.1.

In the real case, it is not possible to know the real idea and the corresponding tag in the client core, and it is also not possible to know what the coefficient between the client and the tag is, so the present embodiment can abstract the model between the client tags as follows. Wherein q represents the predicted trend value of a certain product label (such as atmosphere and cost performance) of a client, and p represents the true trend value of a product relative to a certain label (such as atmosphere and cost performance). If the model of the employee and the product is changed, q represents the predicted familiarity of the employee with the product, and p still represents the true tendency value of the product relative to a certain label.

Push to recommended value

The formula of (1) is:

wherein cf is the serial number of the customer label, if is the serial number of the product label, and k is the total number of the recessive features (labels), and is set to be 10-32;

assuming that the true score of the same type of clients in the training sample data on the product is R_ciThe loss function for which the evaluation model can be derived is

To prevent over-fitting the data in the training samples from degrading the generalization capability of the model, an L2 regularization term control was added,

calculating the partial derivatives of the Loss function to q and p, and controlling the variables

The optimal solution is iteratively solved by using gradient descent,

for convenience of expression, where (t) above p represents the value of a certain feature at a time, (t +1) represents the next time, and t +1 represent the context rather than the power concept, i.e., the next feature value is the present feature value minus the present variation value,

the rule is regular parameter, the experience is set to be 0.01-0.05, beta is learning rate, and the experience is set to be 0.01-0.05;

and (4) solving a customer q row vector and a product p column vector by using supervised learning in machine learning.

In the embodiment, the supervised learning in the machine learning is used for solving the q row vectors of the customers and the p column vectors of the products, and the relations between the customers and the products and the relations between the employees and the products have symmetry in the algorithm, so that only the customer scoring matrix (figure 4) and the employee evaluation matrix (figure 5) need to be replaced to input parameters in the actual operation process. The present embodiment uses the relationship between the customer and the product as an example, that is, step S4 includes: the method comprises the steps of obtaining information of a product and average scores of target customers on the product, compiling a get _ item _ info function, obtaining the information of the product from a mysql database, and putting the information into an item _ dist; compiling a get _ train _ data function to extract a training sample from a customer scoring matrix, and returning a list (customerId, Itemid, label), wherein the customerId is a customer portrait ID, the Itemid is a product ID, and the label is an identifier of a positive sample and a negative sample; model training is carried out on training samples formed by list (customerId, Itemid, label), and column vectors of predicted tendency values of all products and column vectors of true tendency values of all products of a target client are obtained.

The specific implementation process of step S4 is as follows:

step one, writing a get _ item _ info function, acquiring product information from a mysql database, putting the product information into an item _ dist, and outputting the following results:

assuming that the ID of the touch screen is 3, item _ dist [3] output is [ 'touch screen', 'cost performance, hardware, informatization' ];

assuming that the ID of the APP is 1, the item _ dist [1] output is [ 'APP', 'mobile phone is independently installed, can be viewed at any time, and has good user experience' ];

and secondly, extracting training sample data, compiling a get _ train _ data function, extracting training samples from the client rating table, and returning a list such as [ (customerId, Itemid, label1), (customerId2, Itemid2, label2), (customerId3, Itemid3, label3) ], wherein label1, label2 and label3 are identifications of positive and negative samples, and assuming that a sample with a score of more than 3 of a product by a client is taken as a positive sample, label is 1, a sample with a score of less than 3 is taken as a negative sample, label is 0, customerId is a client portrait ID, and Itemid is a product ID. When the training samples are extracted, data of only positive samples or only negative samples in certain types of clients are not added into the final training samples, the data of the types need to be filtered, namely the final training samples comprise the positive samples and the negative samples, and the negative samples are products recommended to the clients but not purchased by the clients, so that the accuracy of model training can be improved. The final list output is as follows:

[(‘24’,‘6’,1),(‘24’,‘1’,1),(‘24’,‘3’,1),(‘24’,‘2’,1),(‘24’,‘1’,1),(‘24’,‘22’,0),(‘24’,‘31’,0)]。

thirdly, writing input parameters of a training model function train (train _ data, K, Alpha, Beta, step) as follows:

train _ data represents training sample data returned by the second step function;

the number of K implicit features is also the length or dimension of the end user vector and the product vector;

alpha, canonical parameter;

beta, learning rate;

step, number of iterations.

The function return comprises the vector of the product for the implicit features: the key is the product ID; itemid, a vector with a value of list and a vector of the client for implicit features, and keys are client image ID, customerid, and a vector with a value of list. Performing iterative training on all training data, if a client or a product needs to be initialized for the first training, performing an initialization function by using random. The partial derivative core code for the customer vector according to the Loss of Loss function of the above formula is as follows:

delta＝label-model_predict(customer_vec[customerid],item_vec[itemid])；

customer_vec[customerid][index]+＝beta*(delta*item_vec[itemid][index]-alpha*customer_vec[customerid][index])；

item_vec[itemid][index]+＝beta*(delta*customer_vec[customerid][index]-alpha*item_vec[itemid][index])；

beta 0.9 (decay of learning rate is required, it is desirable to slow down the change in each iteration)

Specification of model _ predict function model prediction value: and inputting a customer vector and a product vector, and returning the distance between the customer vector and the product vector. The iteration through the above train method finally outputs a structure similar to the following:

for example, the customer ID is equal to 1, and the output of customer _ vec [1] is [0.5323232,0.7123123, -0.8123231. ];

the product ID equals 23 and the output of item _ vec [23] is [0.2356232,0.1125673, -0.9123321.

In step S5, a get _ recommend _ result function is written, the input parameters are a customer vector, a product vector, and a customer ID, and the returned result is a list [ (itemid, score), (itemid1, score1) ], where itemid is the product ID and score is the score, that is, the distance is calculated between the vector corresponding to the customer ID and each product vector, and the recommended result is obtained by sorting according to the distance. The vector distance calculation still uses the model _ predict method, and the final recommended form results are [ ('3', 0.599), ('21', 0.421),. ].

In step S6, the recommendation value of the customer for the product and the recommendation value of the employee for the product are multiplied according to the above algorithm, and when the result of the item is reached, the product with the highest rank can be recommended to the customer by sorting according to the result of the item.

In this way, in the case that the client intention and the demand label are unknown, the possibility of the deal between the product and the client is calculated through setting the label and the historical data, and the product with the highest possibility is recommended to the client, so that even if the recommended product cannot be most suitable for the client, the deal rate and the working efficiency of business personnel can be improved, and meanwhile, the client demand and the staff benefit are considered.

As shown in fig. 6, the present embodiment further provides a product recommendation system based on clients and employees, including:

and the success value calculation module is used for multiplying the recommendation value of each product to the target customer by the recommendation value of the product to the specific staff to obtain the success values of all the products and recommending the product with the highest success value to the target customer.

The product recommendation system of the embodiment recommends the product with the highest possibility to the client by setting the label and calculating the possibility of the transaction between the product and the client through the historical data under the condition that the client intention and the demand label are unknown, so that the transaction rate and the working efficiency of business personnel can be improved even if the recommended product cannot be most suitable for the client, and meanwhile, the client demand and the staff benefit are considered.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A method for product recommendation based on customers and employees, comprising:

2. The customer and employee based product recommendation method of claim 1 wherein the customer label includes a customer nature, industry of interest, and geographic location.

3. The client and employee based product recommendation method of claim 1 wherein employee tags include gender, length of employment, status of subscription and findings.

4. The client and employee based product recommendation method of claim 1 wherein step S4 includes:

compiling a get _ train _ data function to extract a training sample from a customer scoring matrix, and returning a list (custom rId, Itemid, label), wherein the custom Id is a customer portrait ID, the Itemid is a product ID, and the label is an identifier of a positive sample and a negative sample;

5. The customer and employee based product recommendation method of claim 4 wherein the training samples include positive samples and negative samples.

6. The client and employee based product recommendation method of claim 4 wherein the learning rate of model training gradually decays with iteration turns.

7. A product recommendation system based on customers and employees, comprising: