CN109460816B - User behavior prediction method based on deep learning - Google Patents
User behavior prediction method based on deep learning Download PDFInfo
- Publication number
- CN109460816B CN109460816B CN201811365272.6A CN201811365272A CN109460816B CN 109460816 B CN109460816 B CN 109460816B CN 201811365272 A CN201811365272 A CN 201811365272A CN 109460816 B CN109460816 B CN 109460816B
- Authority
- CN
- China
- Prior art keywords
- page
- access
- user
- purchasing
- sequence data
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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"
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a user behavior prediction method based on deep learning, which is characterized by comprising the following steps of: the method comprises the following steps: generating user sequence data; step two: accessing sequence data processing; step three: training a recurrent neural network model; step four: and predicting the purchase probability in real time. The invention can predict the purchasing probability in real time for each step of visiting behavior after the current user enters the website, and can perform marketing intervention in time, thereby improving the effect of website conversion.
Description
Technical Field
The invention relates to the field of user behavior prediction, in particular to a user behavior prediction method based on deep learning.
Background
More and more user data on the internet are recorded, and the action data of searching, clicking, purchasing and the like of the user on a website reflect the interest and the demand of the user on products. Through analysis and modeling of a series of behavior data of the user, the degree of interest of the user in products and websites can be predicted, and therefore data support is provided for marketing of the user, website popularization and the like.
The key problems of insufficient feature expression capability, dimension disaster and other pattern recognition directions are well solved by strong modeling and characterization capabilities of deep learning. The recurrent neural network can solve the problem of serialization correlation, and is a neural network for modeling sequence data, namely the current output of a sequence is also related to the previous output. The specific expression is that the network memorizes the previous information and applies the previous information to the calculation of the current output, and context information can be saved through an intermediate state as an input to influence the prediction of the next time sequence, including but not limited to natural language, speech recognition and serialization labeling problems.
Therefore, the method realizes accurate prediction of the ordering purchase probability of the user as much as possible based on massive online data of the user and by combining a deep learning model. And predicting the purchase possibility and the following behaviors according to the real-time access behavior sequence of the user, and simultaneously carrying out promotion application by combining the characteristic analysis of the user, the browsing dangerous seeds and the like.
The recurrent neural network RNN is a sequence-to-sequence model of the neural network that emphasizes the reuse of a known unit structure.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a user behavior prediction method based on deep learning.
In order to solve the technical problem, the invention provides a user behavior prediction method based on deep learning, which is characterized by comprising the following steps:
the method comprises the following steps: generating user access sequence data, and forming the access sequence data by each purchasing and non-purchasing user according to the access time sequence, wherein the access sequences are the access habits of the users before purchasing products; the access sequence constitutes user sequence data;
step two: and (3) processing the access sequence data generated in the step one, wherein the processing process comprises the following steps:
1) removing unimportant webpage modules, evaluating the importance of the webpage modules, and eliminating webpage modules with the importance lower than the requirement in the access sequence data;
2) the method comprises the steps that a user carries out duplicate removal on continuous same behaviors, and the access sequence data with the continuous same behaviors carries out duplicate removal on the continuous same behaviors of the user;
step three: training a recurrent neural network model;
training a recurrent neural network to predict an access sequence, inputting access sequence data of page access in the step two, outputting the access sequence data of the page access, and if the output access sequence data is insufficient in length, completing the data by using irrelevant marks, and finally adding a Softmax layer before outputting to obtain the purchase probability corresponding to the access sequence data;
step four: forecasting the purchasing probability in real time;
after the user clicks on the website, the possibility that the user may buy the product in the next step, namely the purchase probability, is predicted in real time through the trained recurrent neural network model.
In the second step, webpage modules needing to be removed and having lower importance than requirements comprise purchase behavior pages, namely pages which are necessary before payment is successful, such as a page for adding insurance information, an insurance information confirmation page, a payment page and the like, and pages which are rarely accessed by users in the whole website page structure, wherein the removal method is to count the number of times of access by all users in the latest month of the page, and remove the pages with the access number less than or equal to 3.
The second step further comprises the following steps:
3) further analyzing the data purchased for multiple times, and intercepting access sequence data of multiple purchasing actions for the data purchased for multiple times by taking the purchasing actions as nodes;
4) determining the length of the cut-off maximum access sequence, firstly, filtering the non-purchased access sequence data with the access path less than 5 steps, reserving the access sequence data with the effective access path at least 8 steps, filling with blank spaces with the access path less than 8 steps, selecting the step of pushing forward the purchasing action by 10 steps as the maximum length of the access sequence, and cutting off the access sequence more than 10 steps.
In the third step, the recurrent neural network adopts an LSTM algorithm model, the state is calculated in an accumulation mode, training data are divided into two types, namely purchasing samples and not purchasing samples, and the data volume is more than 50 thousands, particularly up to 100 thousands.
And in the fourth step, when the purchase probability of the user meets the set threshold condition, directionally pushing a marketing promotion scheme for the user according to different service scenes, wherein the marketing promotion scheme comprises promotion means such as coupon issuing, manual service, purchase guidance and the like.
Has the advantages that: according to the method, the online shopping probability prediction is carried out on each step of visiting behavior after the current user enters the website by adopting an LSTM long-and-short-term memory network model based on the sequence data formed by the visiting behavior of the user, so that marketing intervention is carried out in time, and website conversion is improved. The beneficial effects of the invention are mainly embodied in three aspects: (1) the access time sequence data of purchasing and non-purchasing users are adopted for prediction classification, and meanwhile, sample data is specially processed; (2) by adopting a long-time memory network training behavior sequence prediction model in deep learning, the up-and-down walking in the access time sequence data can be captured as an incidence relation, so that the prediction accuracy is greatly improved; (3) after the model is established, for the user who enters the website at present, along with the deepening of the user access behavior, the real-time purchase possibility prediction is carried out after the access of each step, marketing intervention is carried out in time according to the purchase intention, website conversion is improved, and timeliness is greatly improved.
Drawings
FIG. 1 is a simplified flow diagram of an exemplary embodiment of the present invention;
FIG. 2 is a diagram of the predictive effect of purchase probability in accordance with an exemplary embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and the exemplary embodiments:
as shown in fig. 1, a method for predicting user behavior based on deep learning includes the following steps:
s11: user access sequence data generation
The user can generate a plurality of behaviors on the website, some behaviors trigger purchase behaviors, and the order purchase of the user is the final purpose for the e-commerce website. For each purchasing user, acquiring behaviors before the user purchases through the analyzed log, wherein the behaviors have obvious front-back order relation, and each purchasing user can form an access sequence according to the access time sequence, so that the front-back order behaviors are connected to be used as access sequence data before the user purchases, the access sequences are access habits before the user purchases products, and the access sequences form user sequence data; reflecting the desire and need for purchase. The user identity cookie recorded by the website log is adopted to identify different users, a series of click behaviors of the users entering the website are well arranged according to time sequence to form access behavior sequence data, short paths are firstly filtered (a large number of paths, particularly paths not purchased are less than 5 steps), a sequence of at least 8 steps of effective access paths is reserved, and the paths less than 8 steps are filled with blanks. And respectively obtaining purchased sequence samples and unpurchased sequence samples through preprocessing.
Taking the insurance website as an example, there is the following user access sequence S:
the "S" is a special subject page, home page, insurance classification page, comprehensive unexpected insurance list page, member center page, overseas travel insurance list page, global travel insurance products for Annun journey page, global travel comprehensive insurance crown plan products page, overseas travel insurance list page, global travel comprehensive insurance crown plan products page, member login page, global travel comprehensive insurance crown plan products page, insurance information filling page, insurance application confirmation page, successful purchase page, home page, special subject page, domestic travel insurance list page, baby 'S happy disease hospitalization guarantee products page, mom' S happy child 'S disease hospitalization guarantee products page, baby' S happy disease hospitalization guarantee products page, insurance application information filling page, insurance application confirmation page, successful payment page, home page, and claim page
Sequence S has 26 pages visited, 2 purchases.
S12: access sequence data processing
The sequence preprocessing of the step is important, and high-quality training data are provided for subsequent algorithms.
The quality of the sequence data generated in the last step is not high, and the generated sequence needs to be further processed, which mainly comprises the following aspects:
1) removing unimportant web page modules
The web site has a plurality of pages, the importance of each page is different, unimportant modules do not have important influence on final purchasing behavior, but have negative influence on final purchasing prediction, and therefore the pages need to be removed. In the second step, the webpage modules with the importance lower than the requirement to be rejected include purchasing behavior pages, that is, pages necessary before payment succeeds, such as an insurance information adding page, an insurance information confirming page, a payment page, and the like, which cannot correctly reflect the behavior path and purchasing preference of the user, and therefore, the webpage modules need to be rejected.
For example, in the sequence S in the above example, where the insurable page is a web page that needs to be clicked before each purchase, there is no incentive for the purchase, so we eliminate this web page module, and the sequence S becomes:
the page comprises a special subject page, a home page, an insurance classification page, a comprehensive accident insurance list page, a member center page, an overseas travel insurance list page, an all-round trip global insurance product page, a global travel comprehensive guarantee crown plan product page, an overseas travel insurance list page, a global travel comprehensive guarantee crown plan product page, a member login page, a global travel comprehensive guarantee crown plan product page, a purchase success page, a home page, a special subject page, a domestic travel insurance list page, a baby 'S music disease hospitalization guarantee product page, a mother' S music child 'S disease hospitalization guarantee product page, a baby' S music disease hospitalization guarantee product page, a payment success page, a home page and a claim page.
The method also comprises the page which is rarely accessed by the user in the whole website page structure, and the information quantity of the part of the page which reflects the access path and preference of the user is too weak, which can affect the accuracy of model prediction, so that the part of the page needs to be removed. The adopted elimination method is to count the number of times of the page visited by all users in the last month and remove the page with the visiting number less than or equal to 3.
2) User continuous same behavior deduplication
For users with purchase intentions, the users may compare related products horizontally for a long period of time or continuously search for products, such a sequence may have continuous same behaviors, the users continuously access the same page module, and the number of times each user continuously accesses may be greatly different, which may result in an excessively long sequence length, and for the effectiveness of the subsequent algorithm, it is determined to perform deduplication on the continuous same behaviors under comprehensive consideration.
For example, the sequence S is repeated by the same sequence to obtain the following sequence:
the system comprises a special subject page, a home page, an insurance classification page, a comprehensive accident insurance list page, a member center page, an overseas travel insurance list page, an all-round trip global insurance product page, a global travel comprehensive insurance imperial crown plan product page, an overseas travel insurance list page, a global travel comprehensive insurance imperial crown plan product page, a member login page, a global travel comprehensive insurance imperial crown plan product page, a purchase success page, a home page, a special subject page, a domestic travel insurance list page, a baby ' S disease hospitalization guarantee product page, a mother ' S music child disease hospitalization guarantee product page, a baby ' S music disease hospitalization guarantee product page, a payment success page, a home page and a claim page.
3) Multiple purchase intercepts yield multiple pieces of data.
The user may generate multiple purchasing behaviors in one day, the behaviors of the user are different for each purchasing behavior, one purchasing behavior corresponds to one behavior sequence, multiple purchasing behaviors are generated for multiple purchasing, and the multiple purchasing behaviors are taken as multiple sequence data, wherein the multiple purchasing data takes the purchasing behavior as a node to intercept the multiple sequence data of the purchasing behavior.
For the above access sequence S, a total of 2 purchases occur, and then taking the sequence data before the first purchase, such as the above sequence S, the following sequence is obtained:
the page comprises a special subject page, a home page, an insurance classification page, a comprehensive unexpected insurance list page, a member center page, an overseas travel insurance list page, an Annu global travel insurance product page, a global travel comprehensive guarantee crown plan product page, an overseas travel insurance list page, a global travel comprehensive guarantee crown plan product page, a member login page, a global travel comprehensive guarantee crown plan product page and a successful purchase page
4) Truncating the maximum sequence length of 10
The sequence length formed by each user is different, and through analyzing the sequence length before the user purchases, the fact that 10 steps are taken from purchase to the front when the maximum sequence length is 10 and the sequence length is more than 10 is found out for improving the development efficiency of the subsequent algorithm. S is changed to the following sequence
S is an insurance classification page, a comprehensive unexpected insurance list page, a member center page, an overseas travel insurance list page, an ampere-trip global travel insurance products page, a global travel comprehensive guarantee crown plan products page, an overseas travel insurance list page, a global travel comprehensive guarantee crown plan products page, a member login page, a global travel comprehensive guarantee crown plan products page, a successful purchase page
The training data comprises two parts of sample data which are purchased and not purchased, and more than 50 ten thousand, particularly 100 ten thousand data size are respectively acquired as training samples.
User behavior sequence Listing as Table 1
Wherein, the less than 10 steps are filled with blank spaces.
S13: training recurrent neural network model
In the aspect of time sequence prediction, the recurrent neural network has great advantages, and the recurrent neural network RNN has the problems of gradient disappearance and gradient explosion when the sequence is long. While LSTM can avoid the RNN gradient vanishing problem, it uses an additive form to calculate the state, which results in the derivative also being additive, thus avoiding gradient vanishing.
And inputting a path sequence of page access in the step two, outputting the path sequence of page access, and finally outputting the path sequence, wherein the softmax layer is added before the path sequence is output to obtain the purchase probability corresponding to the path.
The input and output are roughly of the form:
input: insurance classification page, integrated unexpected insurance list page, member center page, overseas travel insurance list page, global travel insurance products for Annun journey page, global travel integrated guarantee crown plan products page, overseas travel insurance list page, global travel integrated guarantee crown plan products page, member login page, global travel integrated guarantee crown plan products page, purchase of members, and purchase of members
Output: the comprehensive accident insurance list page, the member center page, the overseas travel insurance list page, the Anlun Global travel insurance product page, the Global travel comprehensive guarantee crown plan product page, the overseas travel insurance list page, the Global travel comprehensive guarantee crown plan product page, the Member login page, the Global travel comprehensive guarantee crown plan product page, the purchase, END.
Output lags Input by one step, and since the length requirement for each Input in batch is consistent during training, short paths are filled with irrelevant flags, such as END.
The following is the implementation process:
s14: purchase probability real-time prediction
After the LSTM model is trained, for any click of a user on a website, the purchase probability of the user after the click can be calculated, and if the purchase probability meets a set threshold condition, sales promotion means such as issuing coupons, manual service, purchase guidance and the like can be triggered to stimulate the consumption of the user. With the deepening of the user access behaviors, the purchase possibility is predicted in real time after the access behaviors of each step, marketing intervention can be performed in time, and website conversion is improved.
Fig. 2 is a diagram showing the predicted effect of each step of the purchase probability obtained by a certain user accessing the following pages on the website.
The invention is mainly used for providing a user behavior prediction method based on deep learning, and the method is based on sequence data formed by user access behaviors, adopts an LSTM long-time memory network model, carries out real-time purchase probability prediction on each step of access behaviors of a current user after the current user enters a website, carries out marketing intervention in time and improves website conversion. The beneficial effects of the invention are mainly embodied in three aspects: (1) the access time sequence data of purchasing and non-purchasing users are adopted for prediction classification, and meanwhile, sample data is specially processed; (2) by adopting a long-time memory network training behavior sequence prediction model in deep learning, the up-and-down walking in the access time sequence data can be captured as an incidence relation, so that the prediction accuracy is greatly improved; (3) after the model is established, for the user who enters the website at present, along with the deepening of the user access behavior, the real-time purchase possibility prediction is carried out after the access of each step, marketing intervention is carried out in time according to the purchase intention, website conversion is improved, and timeliness is greatly improved.
The above embodiments do not limit the present invention in any way, and all other modifications and applications that can be made to the above embodiments in equivalent ways are within the scope of the present invention.
Claims (2)
1. A user behavior prediction method based on deep learning is characterized by comprising the following steps:
the method comprises the following steps: generating user sequence data, namely forming access sequence data for each purchasing user according to the access time sequence, wherein the access sequences are access habits of the users before purchasing products; the access sequence constitutes user sequence data;
step two: and (3) processing the access sequence data generated in the step one, wherein the processing process comprises the following steps: 1) Removing unimportant webpage modules, evaluating the importance of the webpage modules, and eliminating webpage modules with the importance lower than the requirement in the access sequence data; 2) The method comprises the steps that a user carries out duplicate removal on continuous same behaviors, and the duplicate removal is carried out on the continuous same behaviors of the user aiming at access sequence data with the continuous same behaviors;
step three: training a recurrent neural network model;
training a recurrent neural network to predict an access sequence, inputting access sequence data of page access in the step two, and outputting the access sequence data of the page access in the middle, wherein if the length of the output access sequence data is insufficient, the output access sequence data is supplemented by irrelevant marks, and finally, adding a Softmax layer before outputting to obtain the purchase probability corresponding to the access sequence data;
in the third step, the recurrent neural network adopts an LSTM algorithm model, the state is calculated in an accumulation mode, training data are divided into two types, namely purchasing samples and not purchasing samples, and the data volume is more than 50 ten thousand respectively;
step four: forecasting the purchasing probability in real time, namely forecasting the possibility that the user may purchase a product in the next step, namely the purchasing probability, in real time through the trained recurrent neural network model after the user clicks on the website;
in the first step, user identity cookies recorded by website logs are adopted to identify different users, a series of clicking behaviors of the users entering the website are well arranged according to time sequence to form behavior access sequence data, and access sequence data samples of purchasing and not purchasing are obtained, in the second step, webpage modules with the importance lower than the requirement are required to be removed, and the webpage modules comprise purchasing behavior pages, namely pages which are necessary to pass before payment is successful: adding an application information page, an application information confirmation page and a payment page, and also including pages rarely visited by users in the whole website page structure, wherein the adopted elimination method is to count the number of times of visits by all users in the latest month of the page, and remove the pages with the visit number less than or equal to 3;
the second step further comprises the following steps:
3) further analyzing the data purchased for multiple times, and intercepting access sequence data of multiple purchasing actions for the data purchased for multiple times by taking the purchasing actions as nodes; 4) Determining the length of the cut-off maximum access sequence, firstly, filtering the non-purchased access sequence data with the access path less than 5 steps, reserving the access sequence data with the effective access path at least 8 steps, filling with blank spaces with the access path less than 8 steps, selecting the step of pushing forward the purchasing action by 10 steps as the maximum length of the access sequence, and cutting off the access sequence more than 10 steps.
2. The deep learning-based user behavior prediction method of claim 1, wherein: and in the fourth step, when the purchase probability of the user meets the set threshold condition, directionally pushing a marketing promotion scheme for the user according to different service scenes, wherein the marketing promotion scheme comprises issuing coupons, manual service and purchasing guide promotion means.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811365272.6A CN109460816B (en) | 2018-11-16 | 2018-11-16 | User behavior prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811365272.6A CN109460816B (en) | 2018-11-16 | 2018-11-16 | User behavior prediction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109460816A CN109460816A (en) | 2019-03-12 |
| CN109460816B true CN109460816B (en) | 2020-09-18 |
Family
ID=65610702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811365272.6A Active CN109460816B (en) | 2018-11-16 | 2018-11-16 | User behavior prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460816B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110990704A (en) * | 2019-12-06 | 2020-04-10 | 创新奇智(成都)科技有限公司 | Learning prediction method for time series user and content interaction behaviors |
| CN113095861A (en) * | 2020-01-08 | 2021-07-09 | 浙江大搜车软件技术有限公司 | Method, device and equipment for predicting target object transaction probability and storage medium |
| CN113435966B (en) * | 2021-06-22 | 2022-06-03 | 布瑞克农业大数据科技集团有限公司 | Product transaction method and system |
| WO2023097518A1 (en) * | 2021-11-30 | 2023-06-08 | Siemens Aktiengesellschaft | Interface display method and apparatus of industrial software |
| CN115827947B (en) * | 2023-02-03 | 2023-04-25 | 北京匠数科技有限公司 | Method and device for collecting paging website data and electronic equipment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018533148A (en) * | 2015-10-02 | 2018-11-08 | ネイバー コーポレーションＮＡＶＥＲ Ｃｏｒｐｏｒａｔｉｏｎ | Method and system for categorizing data |
Family Cites Families (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106296257A (en) * | 2015-06-11 | 2017-01-04 | 苏宁云商集团股份有限公司 | A kind of fixation of advertisement position put-on method based on user behavior analysis and system |
| CN106485562B (en) * | 2015-09-01 | 2020-12-04 | 苏宁云计算有限公司 | Commodity information recommendation method and system based on user historical behaviors |
| WO2017093953A1 (en) * | 2015-12-02 | 2017-06-08 | Tata Consultancy Services Limited | Method and system for purchase behavior prediction of customers |
| CN105787100A (en) * | 2016-03-18 | 2016-07-20 | 浙江大学 | User session recommendation method based on deep neural network |
| CN107516235B (en) * | 2016-06-17 | 2021-06-15 | 阿里巴巴集团控股有限公司 | Commodity preference estimation method and device |
| US20180046920A1 (en) * | 2016-08-10 | 2018-02-15 | Paypal, Inc. | User Data Learning Based on Recurrent Neural Networks with Long Short Term Memory |
| US10002322B1 (en) * | 2017-04-06 | 2018-06-19 | The Boston Consulting Group, Inc. | Systems and methods for predicting transactions |
| CN107689008A (en) * | 2017-06-09 | 2018-02-13 | 平安科技（深圳）有限公司 | A kind of user insures the method and device of behavior prediction |
| CN108509640A (en) * | 2018-04-11 | 2018-09-07 | 焦点科技股份有限公司 | A kind of page layout optimization method based on sequence prediction |
| CN108563755A (en) * | 2018-04-16 | 2018-09-21 | 辽宁工程技术大学 | A kind of personalized recommendation system and method based on bidirectional circulating neural network |
| CN108648049B (en) * | 2018-05-03 | 2022-03-01 | 中国科学技术大学 | Sequence recommendation method based on user behavior difference modeling |
| CN108648020A (en) * | 2018-05-15 | 2018-10-12 | 携程旅游信息技术（上海）有限公司 | User behavior quantization method, system, equipment and storage medium |
-
2018
- 2018-11-16 CN CN201811365272.6A patent/CN109460816B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018533148A (en) * | 2015-10-02 | 2018-11-08 | ネイバー コーポレーションＮＡＶＥＲ Ｃｏｒｐｏｒａｔｉｏｎ | Method and system for categorizing data |
Non-Patent Citations (1)
| Title |
|---|
| An LSTM-based dynamic customer model for fashion recommendation;Sebastian Heinz 等;《arXiv:1708.07347》;20170824;第1-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109460816A (en) | 2019-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460816B (en) | User behavior prediction method based on deep learning | |
| US10796316B2 (en) | Method and system for identifying fraudulent publisher networks | |
| CN105989004B (en) | Information delivery preprocessing method and device | |
| CN107590169B (en) | Operator gateway data preprocessing method and system | |
| US10574766B2 (en) | Clickstream analysis methods and systems related to determining actionable insights relating to a path to purchase | |
| CN111461841B (en) | Article recommendation method, device, server and storage medium | |
| US8412665B2 (en) | Action prediction and identification temporal user behavior | |
| US20110231256A1 (en) | Automated building of a model for behavioral targeting | |
| CN105184618A (en) | Commodity individual recommendation method for new users and system | |
| CN103797474A (en) | Multi-channel conversion path position reporting | |
| CN105469263A (en) | Commodity recommendation method and device | |
| TW201514889A (en) | Customer data analysis system | |
| CN105335518A (en) | Method and device for generating user preference information | |
| CN108153792A (en) | A kind of data processing method and relevant apparatus | |
| CN113191838A (en) | Shopping recommendation method and system based on heterogeneous graph neural network | |
| Dias et al. | Automating the extraction of static content and dynamic behaviour from e-commerce websites | |
| CN112055038B (en) | Method for generating click rate estimation model and method for predicting click probability | |
| CN103605736A (en) | Method and device for processing conversion data | |
| CN105183878A (en) | Music classification recommending method based on Markov prediction algorithm | |
| Alquhtani et al. | Development of Effective Electronic Customer Relationship Management (ECRM) Model by the Applications of Web Intelligence Analytics | |
| WO2021223165A1 (en) | Systems and methods for object evaluation | |
| CN115794898B (en) | Financial information recommendation method and device, electronic equipment and storage medium | |
| JP5835802B2 (en) | Purchase forecasting apparatus, method, and program | |
| Zhang et al. | Identification of factors predicting clickthrough in Web searching using neural network analysis | |
| CN113722487A (en) | User emotion analysis method, device and equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |