CN112183875A - Multi-factor online purchasing behavior conversion prediction method based on user and product level - Google Patents

Multi-factor online purchasing behavior conversion prediction method based on user and product level Download PDF

Info

Publication number
CN112183875A
CN112183875A CN202011085876.2A CN202011085876A CN112183875A CN 112183875 A CN112183875 A CN 112183875A CN 202011085876 A CN202011085876 A CN 202011085876A CN 112183875 A CN112183875 A CN 112183875A
Authority
CN
China
Prior art keywords
user
product
purchase
probability
time
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
Application number
CN202011085876.2A
Other languages
Chinese (zh)
Inventor
曹杰
张佳禹
申冬琴
赵慕阶
张威煊
徐桂莹
马丽娜
靖慧
罗婕
张兴旺
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.)
Yunjing Business Intelligence Research Institute Nanjing Co ltd
Nanjing University of Science and Technology
Original Assignee
Yunjing Business Intelligence Research Institute Nanjing Co ltd
Nanjing University of Science and Technology
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 Yunjing Business Intelligence Research Institute Nanjing Co ltd, Nanjing University of Science and Technology filed Critical Yunjing Business Intelligence Research Institute Nanjing Co ltd
Priority to CN202011085876.2A priority Critical patent/CN112183875A/en
Publication of CN112183875A publication Critical patent/CN112183875A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0623—Item investigation
    • G06Q30/0625—Directed, with specific intent or strategy
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Abstract

The invention discloses a multi-factor online purchasing behavior conversion prediction method based on user and product level, which is characterized in that the method extracts the prediction factors in the browsing history of a user based on a series of page access events of the user to obtain the purchasing probability of the user caused by each prediction factor; aggregating the obtained user purchase probability by combining the bridge constant of the product level to obtain a user conversion cardinality eta[t0:t0+△t](p)，η[t0:t0+△t](p) represents the probability that user c purchased product p within time Δ t. The prediction method and the prediction results of the 5 reference methods are subjected to error analysis, the RMSE value of the prediction method is lowest, the PCC value is highest, and the method is proved to have higher accuracy.

Description

Multi-factor online purchasing behavior conversion prediction method based on user and product level
Technical Field
The invention relates to the field of electronic commerce, in particular to a multi-factor online purchasing behavior conversion prediction method based on user and product levels.
Background
With the deep development of electronic commerce, the scale of e-commerce websites and e-commerce consumers is in a sharp increase situation, the online purchasing habit of users is rapidly formed, massive product information and online behavior data of the consumers are brought, how to extract and analyze valuable user behavior information from the massive data is achieved, the user preference and the purchasing behavior are accurately predicted, the purchasing conversion rate of products is further improved, and the method is an important aspect of improving the competitiveness of enterprises.
Patent "method for predicting purchase probability based on behavior order of user and apparatus therefor" (patent application No. 201810745481.7), invented a method for predicting user purchase probability and apparatus therefor, which arranges user logs accessing shopping site collected in real time in time order, generates Uniform Resource Identifier (URI) sequence corresponding to user's online behavior, calculates user's product purchase probability by comparing the URI sequence with purchase probability model corresponding to shopping site, and improves accuracy of detecting user's current purchase intention.
The patent "a method and apparatus for predicting user's behavior of purchasing goods" (patent application No. 201610356742.7), provides a scheme for predicting user's behavior of purchasing goods, and this scheme considers the goods feature vector sequence G, the user feature vector U and the time difference factor before the sequence G at the same time, mainly solves the defect of low prediction accuracy existing in the prior art.
The patent 'a user purchase intention prediction method based on mobile big data' (patent application number 201910021407.5), relates to a user purchase intention prediction method based on mobile big data, and comprises the steps of constructing feature engineering according to basic information of users and commodities, establishing a plurality of machine learning models, fusing the models and optimizing model network parameters. A commodity recommendation model for building mobile electronic commerce is constructed through big data and an algorithm, the purchase intention of a user is predicted, and appropriate content is accurately recommended for the mobile user.
The three technical schemes all have the following defects: the accuracy of the online prediction of the conversion of the user purchasing behavior is not high; the accuracy of accurate recommendation of related products is not high.
Disclosure of Invention
In order to solve the problems in the background technology, the invention integrates the shopping behavior diversification of the user, the product interest level and the product type, constructs the mutual dependency relationship between the product and the customer mode conversion, provides a user-product multi-factor combined prediction model, optimally estimates the given product on the basis of improving the conversion rate of browsing customers, and recommends the product according to the product conversion rate.
The technical scheme is as follows:
the invention discloses a multi-factor online purchasing behavior conversion prediction method based on user and product level, which extracts prediction factors in user browsing history to obtain each prediction factor based on a series of page access events of a userMeasuring the probability of user purchase caused by factors; aggregating the obtained user purchase probability by combining the bridge constant of the product level to obtain a user conversion cardinality eta[t0:t0+△t](p)，η[t0:t0+△t](p) represents the probability that user c purchased product p within time Δ t.
Preferably, the predictor includes purchase diversion B1 c,pBrowsing ratio B2 c,pBrowsing time B3 c,pSpecifically, the method comprises the following steps:
purchase transfer: using user transfer weight W to represent user purchase probability
Figure BDA0002720371820000021
G (P, V, W) represents a purchase transition graph of a directed node, P ═ P1，P2…PnDenotes different products, the edge denotes the purchase diversion:
Figure BDA0002720371820000022
Z(c,pi,tv) 1 denotes user c at time tvPurchase product pi(ii) a When Z is 0, it means that the product is not purchased; wpi,pj,△tThe higher the indication, the more the user purchases product piThe greater the likelihood of (a); z (c, p)j,[tv:tv+△t]) Indicating user c at time tvWhether product p was purchased +. DELTA.ti；
Browsing ratio B2 c,pThe factor-induced purchase probability of a user is
Figure BDA0002720371820000023
Figure BDA0002720371820000024
YPRRepresenting the number of times a user accesses a p product, NPRIndicating the number of times the user has not accessed the p-product;
browsing time B3 c,pThe factor-induced user transition probability is
Figure BDA0002720371820000025
The users are divided into two groups, one group is a buyer who successfully purchases, and the other group is a shop window shopper, and the generation is based on two customer groups and the browsing duration xactiveProbability density function xadoptThe estimated transition probability is:
Figure BDA0002720371820000026
YPDindicating the duration of access to a p-page, NPDIndicating the duration of time that no p-page was accessed.
Preferably, the browsing duration xactiveThe calculation method comprises the following steps:
defining (c, p) as a client-product pair, defining three transition states between pairs as { inactive, active, adopt }, each pair (c, p) being initially in the inactive state, which means that client c is not aware of the need for p; next, if the (c, p) pair becomes active status, which means the user's demand for p products, the active status can be recognized when c starts to access the product page; if customer c accesses different pages after (c, p) reaches active state, (c, p) still keeps active state; when customer c makes a purchase order for product p, (c, p) becomes an adopt state; after reaching the scope state, the pair (c, p) returns to the inactive state immediately;
duration of browsing xactiveIs the time from active state to adopt state or the length of time the p page was last accessed.
Preferably, the calculation formula of the user conversion base is as follows:
Figure BDA0002720371820000031
wherein, Cactive(p,t0)A set of clients indicating an active state of p at the current time t 0;
preferably, the user uses v ═ c for a series of page access eventsv,rv,tv,θv) Is represented by cvIndicating the user identity ID, rvIndicating page access ID, tvDenotes access time, θvRepresenting product information; all page access events of a user form a page click stream v according to the time sequencec＝{v1,v2,v3…}，v1<v2When event v occurs in this page, the user is said to have accessed the product, product information is recorded, and a note is made as to whether the product was purchased.
Preferably, the page clickstream uses 9300 million user log logs and clickstream events, inclusive of the RecSys2015 dataset.
Preferably, it further comprises a correction step of the conversion error.
Preferably, the conversion error of the purchase transfer is corrected using a naive regression algorithm:
using xpAs the actual observed user purchase probability, ypEstimating a user conversion base eta using a linear regression equation as the estimated user purchase probability[t0:t0+△t](p)：
Figure BDA0002720371820000037
Figure BDA0002720371820000032
Figure BDA0002720371820000038
Represents the set of customers in the adopt state of p during Δ t time; cactive(p,t0)Is shown at the current time t0-a set of clients with Δ t in active state p;
given any two sets X and Y, the slope beta of the linear regression equation is obtained through learningΔtAnd intercept αΔt：
Figure BDA0002720371820000033
Figure BDA0002720371820000034
Defining a linear regression equation Y ═ alphaΔt+βΔtX, estimating the conversion base of the user:
Figure BDA0002720371820000035
in the formula (I), the compound is shown in the specification,
Figure BDA0002720371820000036
and when i is 1, 2 and 3, the purchase probability of the user caused by different prediction factors is respectively expressed.
Preferably, the periodic perception regression model is used to correct the regression prediction time period: based on the simulated dataset, including RecSys2015 dataset and the non-public e-commerce dataset, the sum of the three procurement factors is effectively re-evaluated by the cycle-aware regression model to predict time periods according to the given product and given properties.
Preferably, the forward co-phasing PA correction method is used to influence the predicted performance at the product level: actual observed user purchase probability is coupled with estimated user purchase probability by a PA correction method based on user log and click stream event data.
The invention has the advantages of
Compared with the C-MLE maximum likelihood estimation method, the C-NBR recommendation system model, the C-AVE page cumulative access method, the P-HCR historical conversion rate model and the P-HCC historical conversion base number method (verified by using the root mean square error RMSE and the Pearson correlation coefficient PCC), the method has the advantages that the accuracy of online prediction of user purchasing behavior conversion and the accuracy of accurate recommendation of related products are greatly improved.
Drawings
FIG. 1 is a schematic illustration of the effect of single and combined predictors on the probability of a transition prediction, wherein:
FIG. 1(a) is a schematic diagram showing the influence of a single "purchase transfer" on the prediction probability of a transfer
FIG. 1(b) is a schematic diagram showing the influence of the combination of "browsing rate + browsing duration" on the conversion prediction probability
FIG. 1(c) is a schematic diagram showing the influence of the combination of "purchase transfer + browsing rate + browsing duration" on the conversion prediction probability
FIG. 2 is a schematic diagram of correcting user and product level prediction model transformation errors, wherein:
FIG. 2(a) is a schematic diagram of the error of the prediction value corrected by naive regression
FIG. 2(b) is a comparison graph of predicted value errors after PA and NA correction
FIG. 3 is a graph showing a comparison of performance of the prediction method and the reference method
FIG. 3(a) is a graphical representation of a comparison of the performance of the prediction method of the present disclosure and the reference method for RMSE validation
FIG. 3(b) is a graphical representation of PCC validation performance comparison of the predictive and baseline methods herein
FIG. 4 is a flow chart of the method of the present invention
Detailed Description
The invention is further illustrated by the following examples, without limiting the scope of the invention:
the multi-factor online purchasing behavior conversion prediction method based on the user and product level comprises the following steps: defining a large-scale user online page access data set format, defining a conversion prediction problem, constructing a conversion prediction model based on a user level and a product level, analyzing the influence of a prediction factor, correcting the conversion error of the prediction model of the user level and the product level, and measuring the superiority of the prediction method by utilizing RMSE and PCC. With reference to fig. 4, the specific implementation method is as follows:
step 1, defining a format of a large-scale user online page access data set: defining a unique customer ID for a given e-commerce website, and summarizing a user operation log; defining user pagesPlane access activity v, v ═ (c)v,rv,tv,θv) The method comprises the steps of containing a user identity ID, wherein the page access ID consists of url and name, and the access time tvAnd information product information theta of optional productvAll the page access events of the user can be linked in sequence according to time sequence to form a page access flow vc＝{v1,v2,v3…}，v1<v2The page of the website related to the product description is called the product page, and when the event v occurs in the page, the user accesses the product, the product information is recorded, and whether the product is purchased or not is marked (0 means not purchased, and 1 means purchased).
Step 2, defining a conversion prediction problem: based on the user data of step 1, two prediction problem conversion candidate pairs and conversion simulations are defined, given a weblog, c represents a finite set of customers and p represents a finite set of products. We define (c, p) as the client-product pair, three transition states between pairs are defined as { inactive, active, adopt }: each pair (c, p) is initially in an inactive state, which means that client c is not aware of the need for p; next, if the (c, p) pair becomes active status, which means the user's demand for p products, the active status can be recognized when c starts to access the product page. When client c accesses a different page (a page that is not p) after (c, p) reaches active state, (c, p) remains active; finally, when c makes a purchase order for p products, then (c, p) becomes the scope state. Upon reaching the adht state, the pair (c, p) immediately returns to the inactive state. Formally, a conversion candidate pair may be defined as a set of customers (denoted C)active(p, t0) is in the active state of p at the current time t 0). Meanwhile, the conversion base is used to simulate the conversion rate problem of the user, and the conversion base can be defined as the probability that the user c purchases the product p within the time delta t.
Step 3, constructing a conversion prediction model based on user and product levels: three predictors are extracted from the user's browsing history: purchase diversion (B)1 c,p) Browsing ratio (B)2 c,p) Duration of browsing(B3 c,p) (ii) a Based on the user's click stream event, given a candidate pair (c, p), a user conversion probability Pr (c, p) is defined as Pr (adopt | v)c) Defining the conversion rate eta as the conversion base number of the user by using the conversion probability of the user[t0:t0+△t](p)。
And 4, analyzing the influence of the prediction factors: analyzing the influence of single or combined three prediction factors of user-level purchase transfer, browsing rate and browsing duration on the conversion probability, wherein the influence of the single and combined prediction factors on the conversion prediction probability is shown in FIG. 1; the purchase probability of the user due to the purchase transition factor is w, and the purchase probability of the user due to the browsing rate factor is
Figure BDA0002720371820000051
The probability of user transition due to browsing time factor is
Figure BDA0002720371820000052
Finally, aggregating the factors by using the bridge constant of the product layer;
purchase transfer: such a transition may determine a purchase demand between purchasing the same or different products, G (P, V, W) representing a purchase transition graph of a directed node, P ═ P { (P) }1，P2…PnAnd the sum of the edges represents purchase transfer, and the user transfer weight is adopted to represent the purchase probability of the user:
Figure BDA0002720371820000053
Z(c,pi,tv) 1 denotes user c at time tvPurchase product pi(ii) a When Z is 0, it means that the product is not purchased; wpi,pj,△tThe higher the indication, the more the user purchases product piThe greater the likelihood of (a); z (c, p)j,[tv:tv+△t]) Indicating user c at time tvWhether product p was purchased +. DELTA.ti；
Browsing ratio: the accumulated visit amount of the user page can be used for obtaining products which are interested by the user, the conversion possibility is increased along with the increase of the page visit amount, and the visit times are in direct proportion to the conversion ratio.
Quantifying the cumulative access effect pair (c, p) of the conversion candidates, estimating the conversion ratio:
Figure BDA0002720371820000061
YPRrepresenting the number of times a user accesses a p product, NPRIndicating the number of times the user has not accessed the p-product;
browsing duration: dividing users into two groups, one group is a buyer who successfully purchases, the other group is a shop window shopper, comparing the time length of the two accessing p pages (the time from active state to adopt state or the time length of the last accessing p pages), generating a random variable of browsing time and a probability density function, and estimating the transition probability of a given pair (c, p) based on the probability density functions of two customer groups and browsing duration:
Figure BDA0002720371820000062
YPDindicating the duration of access to a p-page, NPDIndicating the duration of time that no p-page was accessed.
The product factors are aggregated, and the calculation formula for obtaining the user conversion base number is as follows:
Figure BDA0002720371820000063
step 5, correcting the conversion error of the prediction model of the user and product level;
naive regression: to correct for errors based on user-level purchase migration, logistic regression is used to capture more of the user's historical behavior, using x as the actual observed conversion base, ypAs the actual observed conversion base number, the conversion rate was estimated using a linear regression equation:
Figure BDA0002720371820000064
Figure BDA0002720371820000065
periodic perceptual regression: for any one time Δ t, the linear regression in the ideal state varies approximately uniformly over different time periods, and the time period predicted from a given product and a given property by using the sum of the three procurement factors can be effectively reevaluated using cycle-aware regression.
The PA and NA correction method is used for influencing the prediction performance of the product level, and in a more preferred embodiment, the NA correction method is selected for influencing the prediction performance of the product level.
The beneficial effects are verified, and the RMSE and PCC are used for measuring the superiority of the prediction method;
the experimental data set comprises log logs of 9300 ten thousand users including a RecSys2015 data set and click stream events of the users;
setting experimental parameters: the time period Δ t ∈ {7,14,21,30,45,60 }.
The influence of the single and the combination of the prediction factors on the conversion rate is shown in FIG. 1, and the results show that the single and the combination of the prediction factors have positive influence on the conversion prediction probability, the PCC of the combination of the prediction factors is the highest, and the actual conversion base number and the estimated conversion base number of the product are matched with each other more benchmark, which shows that the method of the invention has good performance.
FIG. 2(a) shows that the error of the predicted value after the naive regression correction has a lower RMSE and a better effect; fig. 2(b) shows the predicted value error corrected by PA and NA methods, and the correction effect of PA is better than that of NA, and the PA correction method can better couple the actually observed user purchase probability with the estimated user purchase probability.
Fig. 3 shows the error analysis of the prediction results of the prediction method of the present invention and 5 reference methods, and the prediction method of the present invention has the lowest RMSE value and the highest PCC value, indicating that the method has higher accuracy.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims (10)

1. A multi-factor online purchasing behavior conversion prediction method based on user and product level is characterized in that prediction factors in user browsing history are extracted based on a series of page access events of a user, and user purchasing probability caused by each prediction factor is obtained; aggregating the obtained user purchase probability by combining the bridge constant of the product level to obtain a user conversion cardinality eta[t0:t0+△t](p)，η[t0:t0+△t](p) represents the probability that user c purchased product p within time Δ t.
2. The method of claim 1, wherein the predictor comprises purchase diversion B1 c,pBrowsing ratio B2 c,pBrowsing time B3 c,pSpecifically, the method comprises the following steps:
purchase transfer: using user transfer weight W to represent user purchase probability
Figure FDA0002720371810000011
G (P, V, W) represents a purchase transition graph of a directed node, P ═ P1，P2…PnDenotes different products, the edge denotes the purchase diversion:
Figure FDA0002720371810000012
Z(c,pi,tv) 1 denotes user c at time tvPurchase product pi(ii) a When Z is 0, it means that the product is not purchased; wpi,pj,△tThe higher the indication, the more the user purchases product piThe greater the likelihood of (a); z (c, p)j,[tv:tv+△t]) Indicating that user c is at timetvWhether product p was purchased +. DELTA.ti；
Browsing ratio B2 c,pThe factor-induced purchase probability of a user is
Figure FDA0002720371810000013
Figure FDA0002720371810000014
YPRRepresenting the number of times a user accesses a p product, NPRIndicating the number of times the user has not accessed the p-product;
browsing time B3 c,pThe factor-induced user transition probability is
Figure FDA0002720371810000015
The users are divided into two groups, one group is a buyer who successfully purchases, and the other group is a shop window shopper, and the generation is based on two customer groups and the browsing duration xactiveProbability density function xadoptThe estimated transition probability is:
Figure FDA0002720371810000016
YPDindicating the duration of access to a p-page, NPDIndicating the duration of time that no p-page was accessed.
3. Method according to claim 2, characterized in that the browsing duration xactiveThe calculation method comprises the following steps:
defining (c, p) as a client-product pair, defining three transition states between pairs as { inactive, active, adopt }, each pair (c, p) being initially in the inactive state, which means that client c is not aware of the need for p; next, if the (c, p) pair becomes active status, which means the user's demand for p products, the active status can be recognized when c starts to access the product page; if customer c accesses different pages after (c, p) reaches active state, (c, p) still keeps active state; when customer c makes a purchase order for product p, (c, p) becomes an adopt state; after reaching the scope state, the pair (c, p) returns to the inactive state immediately;
duration of browsing xactiveIs the time from active state to adopt state or the length of time the p page was last accessed.
4. The method of claim 3, wherein the user conversion base is calculated by the formula:
Figure FDA0002720371810000017
wherein, Cactive(p,t0)Indicating a set of clients in active state p at the current time t 0.
5. The method of claim 1, wherein the user's series of page access events are defined by v ═ (c)v,rv,tv,θv) Is represented by cvIndicating the user identity ID, rvIndicating page access ID, tvDenotes access time, θvRepresenting product information; all page access events of a user form a page click stream v according to the time sequencec＝{v1,v2,v3…}，v1<v2When event v occurs in this page, the user is said to have accessed the product, product information is recorded, and a note is made as to whether the product was purchased.
6. The method of claim 5, wherein the page clickstream uses 9300 million user log logs and clickstream events within the RecSys2015 dataset.
7. The method of claim 2, further comprising the step of correcting for conversion errors.
8. The method according to claim 7, characterized in that the conversion error of the purchase transfer is corrected using a naive regression algorithm:
using xpAs the actual observed user purchase probability, ypEstimating a user conversion base eta using a linear regression equation as the estimated user purchase probability[t0:t0+△t](p)：
Figure FDA0002720371810000021
Figure FDA0002720371810000022
Figure FDA0002720371810000023
Represents the set of customers in the adopt state of p during Δ t time; cactive(p,t0)Is shown at the current time t0-a set of clients with Δ t in active state p;
given any two sets X and Y, the slope beta of the linear regression equation is obtained through learningΔtAnd intercept αΔt：
Figure FDA0002720371810000024
Figure FDA0002720371810000025
Defining a linear regression equation Y ═ alphaΔt+βΔtX, estimating the conversion base of the user:
Figure FDA0002720371810000026
in the formula (I), the compound is shown in the specification,
Figure FDA0002720371810000027
and when i is 1, 2 and 3, the purchase probability of the user caused by different prediction factors is respectively expressed.
9. The method of claim 8, wherein the regression prediction time period is modified using a cycle-aware regression model: based on the simulated dataset, including RecSys2015 dataset and the non-public e-commerce dataset, the sum of the three procurement factors is effectively re-evaluated by the cycle-aware regression model to predict time periods according to the given product and given properties.
10. The method of claim 8, wherein the forward synergistic PA correction method is used to affect the predicted performance at the product level: actual observed user purchase probability is coupled with estimated user purchase probability by a PA correction method based on user log and click stream event data.
CN202011085876.2A 2020-10-12 2020-10-12 Multi-factor online purchasing behavior conversion prediction method based on user and product level Pending CN112183875A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011085876.2A CN112183875A (en) 2020-10-12 2020-10-12 Multi-factor online purchasing behavior conversion prediction method based on user and product level

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011085876.2A CN112183875A (en) 2020-10-12 2020-10-12 Multi-factor online purchasing behavior conversion prediction method based on user and product level

Publications (1)

Publication Number Publication Date
CN112183875A true CN112183875A (en) 2021-01-05

Family

ID=73949434

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011085876.2A Pending CN112183875A (en) 2020-10-12 2020-10-12 Multi-factor online purchasing behavior conversion prediction method based on user and product level

Country Status (1)

Country Link
CN (1) CN112183875A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116843388A (en) * 2023-08-29 2023-10-03 新义互联(北京)科技有限公司 Advertisement delivery analysis method and system
CN117217852A (en) * 2023-08-03 2023-12-12 广州兴趣岛信息科技有限公司 Behavior recognition-based purchase willingness prediction method and device

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050049907A1 (en) * 2003-08-26 2005-03-03 Suresh Kumar Using page-view data to project demand for an item
JP2013073534A (en) * 2011-09-29 2013-04-22 Nippon Telegr & Teleph Corp <Ntt> Customer survival rate estimation system, customer survival rate estimation method, customer survival rate estimation processing server and program for customer survival rate estimation processing server
US20140222503A1 (en) * 2013-02-07 2014-08-07 24/7 Customer, Inc. Dynamic prediction of online shopper's intent using a combination of prediction models
CN106056241A (en) * 2016-05-26 2016-10-26 中国太平洋保险（集团）股份有限公司 Control method and device for predicting user behavior based on large data
US20170337572A1 (en) * 2016-05-23 2017-11-23 Adobe Systems Incorporated Sales Forecasting using Browsing Ratios and Browsing Durations
CN109003114A (en) * 2018-06-01 2018-12-14 四川斐讯信息技术有限公司 A kind of marketing strategy prediction technique, system, computer equipment and storage medium
CN109242524A (en) * 2017-07-10 2019-01-18 Sk普兰尼特有限公司 The method and its equipment of behavior sequential prediction purchase probability based on user
CN109741112A (en) * 2019-01-10 2019-05-10 博拉网络股份有限公司 A kind of user's purchase intention prediction technique based on mobile big data
US10387934B1 (en) * 2014-06-12 2019-08-20 Amazon Technologies, Inc. Method medium and system for category prediction for a changed shopping mission

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050049907A1 (en) * 2003-08-26 2005-03-03 Suresh Kumar Using page-view data to project demand for an item
JP2013073534A (en) * 2011-09-29 2013-04-22 Nippon Telegr & Teleph Corp <Ntt> Customer survival rate estimation system, customer survival rate estimation method, customer survival rate estimation processing server and program for customer survival rate estimation processing server
US20140222503A1 (en) * 2013-02-07 2014-08-07 24/7 Customer, Inc. Dynamic prediction of online shopper's intent using a combination of prediction models
US10387934B1 (en) * 2014-06-12 2019-08-20 Amazon Technologies, Inc. Method medium and system for category prediction for a changed shopping mission
US20170337572A1 (en) * 2016-05-23 2017-11-23 Adobe Systems Incorporated Sales Forecasting using Browsing Ratios and Browsing Durations
CN106056241A (en) * 2016-05-26 2016-10-26 中国太平洋保险（集团）股份有限公司 Control method and device for predicting user behavior based on large data
CN109242524A (en) * 2017-07-10 2019-01-18 Sk普兰尼特有限公司 The method and its equipment of behavior sequential prediction purchase probability based on user
CN109003114A (en) * 2018-06-01 2018-12-14 四川斐讯信息技术有限公司 A kind of marketing strategy prediction technique, system, computer equipment and storage medium
CN109741112A (en) * 2019-01-10 2019-05-10 博拉网络股份有限公司 A kind of user's purchase intention prediction technique based on mobile big data

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117217852A (en) * 2023-08-03 2023-12-12 广州兴趣岛信息科技有限公司 Behavior recognition-based purchase willingness prediction method and device
CN117217852B (en) * 2023-08-03 2024-02-27 广州兴趣岛信息科技有限公司 Behavior recognition-based purchase willingness prediction method and device
CN116843388A (en) * 2023-08-29 2023-10-03 新义互联(北京)科技有限公司 Advertisement delivery analysis method and system
CN116843388B (en) * 2023-08-29 2023-11-17 新义互联(北京)科技有限公司 Advertisement delivery analysis method and system

Similar Documents

Publication Publication Date Title
CN102385601B (en) A kind of recommend method of product information and system
McMahan et al. Ad click prediction: a view from the trenches
JP6152173B2 (en) Ranking product search results
JP5175903B2 (en) Adaptive analysis multidimensional processing system
US20230410135A1 (en) System and method for selecting promotional products for retail
CN112131480B (en) Personalized commodity recommendation method and system based on multilayer heterogeneous attribute network representation learning
CN105976229A (en) Collaborative filtering algorithm based on user and project mixing
JP5068382B1 (en) Prediction device, prediction method, and prediction program
CN103399858A (en) Socialization collaborative filtering recommendation method based on trust
BRPI0620830A2 (en) estimated ad quality system and method of use for computer filtering, classification and promotion and computer readable medium
US8954910B1 (en) Device mismatch contribution computation with nonlinear effects
US11854022B2 (en) Proactively predicting transaction dates based on sparse transaction data
CN108205775A (en) The recommendation method, apparatus and client of a kind of business object
CN112183875A (en) Multi-factor online purchasing behavior conversion prediction method based on user and product level
Kim et al. Tracking temporal dynamics of purchase decisions via hierarchical time-rescaling model
CN105701681A (en) Method and device used for predicting order amount
Mendes et al. The application of case-based reasoning to early Web project cost estimation
CN105761154A (en) Socialized recommendation method and device
CN102163304A (en) Method and system for collaborative networking with optimized inter-domain information quality assessment
US8813009B1 (en) Computing device mismatch variation contributions
Wang et al. Adaptive time series prediction and recommendation
Davoudi et al. Modeling trust for rating prediction in recommender systems
Song Web service reliability prediction based on machine learning
CN110083518B (en) AdaBoost-Elman-based virtual machine software aging prediction method
JP2009244981A (en) Analysis apparatus, analysis method, and analysis program

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