CN109460816A - A kind of user's behavior prediction method based on deep learning - Google Patents
A kind of user's behavior prediction method based on deep learning Download PDFInfo
- Publication number
- CN109460816A CN109460816A CN201811365272.6A CN201811365272A CN109460816A CN 109460816 A CN109460816 A CN 109460816A CN 201811365272 A CN201811365272 A CN 201811365272A CN 109460816 A CN109460816 A CN 109460816A
- Authority
- CN
- China
- Prior art keywords
- user
- page
- access
- purchase
- behavior
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- 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
本发明公开了一种基于深度学习的用户行为预测方法，其特征在于，包括如下步骤：步骤一：用户序列数据生成；步骤二：访问序列数据处理；步骤三：训练循环神经网络模型；步骤四：购买概率实时预测。本发明达到对当前用户进入网站后的每一步访问行为进行实时的购买概率预测，及时的进行营销干预，提高网站转化的效果。
The invention discloses a user behavior prediction method based on deep learning, which is characterized by comprising the following steps: step 1: user sequence data generation; step 2: access sequence data processing; step 3: training a cyclic neural network model; step 4 : Real-time prediction of purchase probability. The present invention achieves real-time purchase probability prediction for each step of access behavior after the current user enters the website, timely conducts marketing intervention, and improves the effect of website conversion.
Description
Technical field
The present invention relates to user's behavior prediction fields, more particularly to a kind of user's behavior prediction side based on deep learning
Method.
Background technique
More and more users' data are recorded on internet, the row such as search, click, the purchase of user on website
For data, reflect user to the interest and demand of product.It, can be with by the analysis and modeling to a series of behavioral datas of user
User is predicted to the interested degree of product and website, to market for user, website promotion etc. provides data supporting.
Deep learning powerful modeling and characterization ability very good solution feature representation scarce capacity and dimension disaster etc.
The critical issue in pattern-recognition direction.Recognition with Recurrent Neural Network is able to solve serializing relevant issues, is that a kind of pair of sequence data is built
The output of the neural network of mould, i.e. a sequence current output and front is also related.The specific form of expression is that network can be right
The information of front is remembered and is applied in the calculating currently exported, can save contextual information by intermediate state,
As the prediction of a timing under the influence of input, including but not limited to natural language, speech recognition, serializing mark problem.
Therefore, data on the line herein based on user's magnanimity, it is real to purchase probability single under user in conjunction with deep learning model
Now accurately prediction as far as possible.According to user's real time access behavior sequence, prediction purchase possibility and next behavior, simultaneously
In conjunction with the signature analysis of user, the insurance kind etc. of browsing carries out promotion application.
Recognition with Recurrent Neural Network RNN is model of one sequence of neural network to sequence, emphasizes a known units structure weight
It is multiple to use.
Summary of the invention
The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a kind of use based on deep learning
Family behavior prediction method.
In order to solve the above technical problems, the present invention provides a kind of user's behavior prediction method based on deep learning, it is special
Sign is, includes the following steps:
Step 1: user's access sequence data generate, and do not buy each purchase and user according to access time sequence
An access sequence data are formed, these access sequences are exactly that user buys the access habits before product；Access sequence, which is constituted, to be used
Family sequence data；
Step 2: the access sequence data generated in step 1 are further processed in access sequence data processing, place
Reason process includes the following steps:
1) unessential Web page module is removed, importance sequenencing is carried out to Web page module, to importance lower than desired net
Page module needs are first rejected in access sequence data；
2) the consecutive identical behavior duplicate removal of user, for the access sequence data with continuous identical behavior, to wherein using
The continuous identical behavior at family carries out duplicate removal；
Step 3: training Recognition with Recurrent Neural Network model；
Training Recognition with Recurrent Neural Network accesses sequence prediction, and input is the access sequence number of page access in step 2
According to output is also the access sequence data of page access, and the access sequence data length of such as output is insufficient, then with incoherent mark
Remember polishing, is added Softmax layers before final output and obtains the corresponding purchase probability of access sequence data；
Step 4: purchase probability is predicted in real time；
After user has click behavior on website, by trained Recognition with Recurrent Neural Network model, in real time
Predict a possibility that user is likely to purchase product in next step, i.e. purchase probability.
In the step 1, go to identify different users, Yong Hujin using the user identity Cookie that web log file records
A series of access behaviors for entering website, arrange according to chronological order, form access sequence data, obtain purchase and do not purchase
The access sequence data sample bought, in the step 2, the importance for needing to reject includes purchase lower than desired Web page module
The behavior page, that is, the necessary page before paying successfully such as dose information page of insuring, the validation of information page of insuring, payment page
Face etc., further includes the page being seldom accessed by the user in entire Website page structure, and the elimination method of use is to count the page most
The number accessed in nearly one month by all users, the page by access times less than or equal to 3 remove.
Further include following steps in the step 2:
3) data repeatedly bought further are analyzed, to the data repeatedly bought with buy movement be node intercept out it is a plurality of
The access sequence data of purchase movement；
4) length that maximum access sequence is truncated is determined, firstly, the access sequence do not bought of the filtered access path less than 5 steps
Column data retains the access sequence data of effective access path at least 8 steps, uses space polishing less than 8 steps, chooses purchase movement
Toward maximum length of 10 steps as access sequence is pushed forward, the access sequence greater than 10 is truncated.
In the step 3, Recognition with Recurrent Neural Network uses LSTM algorithm model, uses cumulative form calculus state, training
Data are divided into two classes, respectively buy sample and do not buy sample, and data volume is respectively 500,000 or more, and especially up to 100
Ten thousand.
In the step 4, when the purchase probability of user meets the threshold condition of setting, it is according to different business scene
User orients push promotion scheme, and the promotion scheme includes providing discount coupon, manual service, purchase guidance etc. to promote
Pin means.
The utility model has the advantages that the present invention is based on the sequence data that user access activity is formed, using LSTM long memory network in short-term
Model, each step access behavior after website is entered to active user carry out real-time purchase probability prediction, are timely sought
Pin is intervened, and website conversion is improved.The beneficial effects are mainly reflected as follows three aspects: (1) being used using purchase and non-purchase
The access time sequence data at family carries out prediction classification, while having carried out specially treated to sample data；(2) deep learning is used
In the length behavior sequence prediction model of memory network training in short-term, walking up and down in accessing time sequence data can be captured to close
Connection relationship greatly improves the accuracy of prediction；(3) after model foundation, to the user for being currently entering website, as user accesses
Behavior is goed deep into, and predicts purchase possibility in real time is all carried out after the access of each step, timely according to the wish size of purchase
Carry out marketing intervention, improve website conversion, greatly improve timeliness.
Detailed description of the invention
Fig. 1 is the general flow chart of exemplary embodiment of the present invention；
Fig. 2 is the purchase probability prediction effect figure of exemplary embodiment of the present invention.
Specific embodiment
The present invention is further illustrated with exemplary embodiment with reference to the accompanying drawing:
As shown in Figure 1, a kind of user's behavior prediction method based on deep learning includes the following steps:
S11: user's access sequence data generate
User can generate many behaviors on website, and some behaviors trigger buying behavior, and for electric business website
It is final purpose that user, which places an order and buys,.User is bought for each, is purchased by the way that the log parsed is available to user
Pervious behavior is bought, these behaviors are that have apparent succession relationship, each purchase user is according to access time sequence
An access sequence can be formed, we have the behavior of succession to connect as before user's purchase these thus
Access sequence data, these access sequences are exactly that user buys the access habits before product, and access sequence constitutes user's sequence number
According to；Reflect the hope and demand of purchase.Here we go identification different using the user identity cookie of web log file record
User, user enters a series of click behaviors of website, arranged according to chronological order, forms access behavior sequence number
According to filtering shorter path (path that a large amount of paths are not bought especially is less than 5 steps) first, retain effective access path herein
At least sequence of 8 steps uses space polishing less than 8 steps.The sequence samples for obtaining purchase by pretreatment respectively and not buying.
For insuring website, such as just like next user's access sequence S:
S=special topic page, homepage, insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance
List page, the dangerous product page of peace way whole world travelling, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list
Page, overseas travel insurance list page, worldwide travel comprehensive coverage imperial crown planned product page, member's log in page, member's log in page, ring
Ball travelling comprehensive coverage imperial crown planned product page, information solicitation page of insuring, confirmation page of insuring buy successfully page, homepage, special topic
Page, home journey insure list page, and Baobaole disease, which is hospitalized, ensures product page, and the happy children disease of mother, which is hospitalized, ensures product page,
Baobaole disease, which is hospitalized, ensures product page, information solicitation page of insuring, and confirmation page of insuring pays successfully page, homepage, page of settling a claim
Sequence S shares 26 accession pages, 2 purchases.
S12: access sequence data processing
The pre- data processing of this step sequence is important, provides the training data of high quality for subsequent algorithm.
The sequence data that previous step generates is of low quality, needs to further process the sequence of generation, mainly includes
It is following aspects:
1) unessential Web page module is removed
The page of website is various, and the importance of each page is also different, and unessential module is to final buying behavior
Important influence will not be generated, but final purchase can be predicted to have a negative impact, need to reject thus.The step 2
In, the importance for needing to reject includes the buying behavior page lower than desired Web page module, that is, necessary page before paying successfully
Face, such as information page of insuring, the validation of information page of insuring, the payment page are dosed, these necessary pages can not be correctly anti-
Using the behavior path at family and the preference of purchase, it is therefore desirable to reject.
For example the sequence S in upper example, China National Investment & Guaranty Corp.'s page require the webpage clicked before being each purchase, not to purchase
Stimulation behavior can be generated, we weed out this Web page module thus, and sequence S becomes:
S=special topic page, homepage, insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance
List page, the dangerous product page of peace way whole world travelling, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list
Page, overseas travel insurance list page, worldwide travel comprehensive coverage imperial crown planned product page, member's log in page, member's log in page, ring
Ball travelling comprehensive coverage imperial crown planned product page buys successfully page, homepage, thematic page, home journey insurance list page, Baobaole
Disease, which is hospitalized, ensures product page, and the happy children disease of mother, which is hospitalized, ensures product page, and Baobaole disease, which is hospitalized, ensures product page, payment
Success page, homepage, page of settling a claim.
It further include the page being seldom accessed by the user in entire Website page structure, this partial page is for reflecting user
The information content of access path and preference is too weak, can have an impact to the precision of model prediction, it is therefore desirable to remove this partial page
Fall.The elimination method of use is to count the number accessed in the page nearest one month by all users, access times are less than etc.
The page in 3 removes.
2) the consecutive identical behavior duplicate removal of user
For the user for having purchase intention, user may be related in lateral comparison in very long following period of time
Product, or product search is constantly carried out, such sequence just has continuous identical behavior, and user's connected reference is same
Page module, and the number of each user's connected reference can also be very different, it is so to will cause sequence length mistake
Long, for the validity of subsequent algorithm, comprehensive consideration, which is made decision, carries out duplicate removal to continuous identical behavior.
Such as above-mentioned sequence S, following sequence is obtained by continuing identical behavior duplicate removal:
S=special topic page, homepage, insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance
List page, the dangerous product page of peace way whole world travelling, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list
Page, worldwide travel comprehensive coverage imperial crown planned product page, member's log in page, worldwide travel comprehensive coverage imperial crown planned product page,
Successfully page is bought, homepage, thematic page, home journey insurance list page, Baobaole disease, which is hospitalized, ensures product page, and mother's laughable matter is virgin
Disease, which is hospitalized, ensures product page, and Baobaole disease, which is hospitalized, ensures product page, pays successfully page, homepage, page of settling a claim.
3) repeatedly purchase interception generates a plurality of data.
User may generate the behavior repeatedly bought within one day, for the behavior bought each time, the row of user
For be all it is different, a buying behavior will correspond to a behavior sequence, thus repeatedly purchase will a plurality of sequence of behavior
Data are herein the sequence data that node intercepts out a plurality of purchase movement to buy movement to the data repeatedly bought.
For access sequence S above, 2 buying behaviors have occurred altogether, then taking the sequence number before initial purchase
According to, such as above-mentioned sequence S, obtain following sequence:
S=special topic page, homepage, insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance
List page, the dangerous product page of peace way whole world travelling, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list
Page, worldwide travel comprehensive coverage imperial crown planned product page, member's log in page, worldwide travel comprehensive coverage imperial crown planned product page,
Buy successfully page
4) maximal sequence length 10 is truncated
The sequence length that each user is formed is different, by the sequence length before analysis user's purchase, finds in order to rear
The promotion of continuous algorithm development efficiency, choosing maximal sequence length here is 10, sequence length taking forward from purchase greater than 10
10 steps.S becomes following sequence
S=insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance list page, peace way are complete
The dangerous product page of ball travelling, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list page, worldwide travel are comprehensive
Ensure that imperial crown planned product page, member's log in page, worldwide travel comprehensive coverage imperial crown planned product page buy successfully page
Training data includes purchase and does not buy two parts sample data, obtains 500,000 or more, especially up to 100 respectively
Ten thousand data volumes are as training sample.
Such as 1 user behavior sequence table of table
Wherein, space polishing is used less than 10 steps.
S13: training Recognition with Recurrent Neural Network model
In terms of time series forecasting, Recognition with Recurrent Neural Network has a very big advantage, Recognition with Recurrent Neural Network RNN sequence compared with
It will appear gradient when long to disappear and gradient explosion issues.And LSTM is avoided that the gradient disappearance problem of RNN, uses cumulative shape
Formula calculates state, and this cumulative fashion causes derivative to be also cumulative fashion, therefore avoids gradient disappearance.
Input is the path sequence of page access in step 2, and output is also the path sequence of page access, final output
It is added softmax layers before and obtains the corresponding purchase probability in path.
Input and output are approximately as form:
Input: insurance classification page, comprehensive casualty insurance list page, member's central page, overseas travel insurance list page, pacifies
The way whole world is travelled dangerous product page, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list page, worldwide travel
Comprehensive coverage imperial crown planned product page, member's log in page, worldwide travel comprehensive coverage imperial crown planned product page, purchase
Output: comprehensive casualty insurance list page, member's central page, overseas travel insurance list page, the peace way whole world are travelled
Dangerous product page, worldwide travel comprehensive coverage imperial crown planned product page, overseas travel insurance list page, worldwide travel comprehensive coverage emperor
It is preced with planned product page, member's log in page, worldwide travel comprehensive coverage imperial crown planned product page, purchase, END.
Output is that Input lags a step, when due to training, for the length requirement one of each input in batch
It causes, so short path incoherent label polishing, such as END.
The following are the processes of realization:
S14: purchase probability is predicted in real time
After having trained LSTM model, for any one click of the user on website, user can be calculated and existed
The size of purchase probability after specifically clicking, if purchase probability meets the threshold condition of setting, so that it may triggering granting discount coupon,
The marketing tools such as manual service, purchase guidance, stimulate customer consumption.Visit with going deep into for user access activity, to each step
Purchase possibility prediction in real time is all carried out after asking behavior, marketing intervention can be timely carried out, improve website conversion.
Fig. 2 is access of some user on website such as lower page, respectively obtains the prediction effect of each step purchase probability
Fruit figure.
Present invention is mainly used for a kind of user's behavior prediction method based on deep learning is provided, the present invention is based on user's visits
Ask the sequence data that behavior is formed, each step using LSTM long memory network model in short-term, after website is entered to active user
Access behavior carries out real-time purchase probability prediction, timely carries out marketing intervention, improves website conversion.Beneficial effect of the invention
Fruit is mainly reflected in three aspects: (1) carrying out prediction classification using the access time sequence data of purchase and non-purchase user, together
When specially treated has been carried out to sample data；(2) using the behavior sequence prediction of memory network training in short-term of the length in deep learning
Model, can capture the walking up and down in accessing time sequence data is incidence relation, greatly improves the accuracy of prediction；(3)
After model foundation, to the user for being currently entering website, with going deep into for user access activity, to all being carried out after the access of each step
Purchase possibility prediction in real time timely carries out marketing intervention according to the wish size of purchase, improves website conversion, greatly
Improve timeliness.
Above embodiments do not limit the present invention in any way, all to be made in a manner of equivalent transformation to above embodiments
Other improvement and application, belong to protection scope of the present invention.
Claims (5)
1. a kind of user's behavior prediction method based on deep learning, which comprises the steps of:
Step 1: user's sequence data generates, each purchase user is sequentially formed an access sequence according to access time
Data, these access sequences are exactly that user buys the access habits before product；Access sequence constitutes user's sequence data；
Step 2: the access sequence data generated in step 1 are further processed in access sequence data processing, process
Journey includes the following steps:
1) unessential Web page module is removed, importance sequenencing is carried out to Web page module, to importance lower than desired webpage mould
Block needs are first rejected in access sequence data；
2) the consecutive identical behavior duplicate removal of user, for the access sequence data with continuous identical behavior, to wherein user's
Continuous identical behavior carries out duplicate removal；
Step 3: training Recognition with Recurrent Neural Network model；
Training Recognition with Recurrent Neural Network accesses sequence prediction, and input is the access sequence data of page access in step 2, defeated
Out and the access sequence data of page access, the access sequence data length deficiency of such as output are then mended with incoherent label
Together, Softmax layers are added before final output and obtain the corresponding purchase probability of access sequence data；
Step 4: purchase probability predicts in real time,
After user has click behavior on website, by trained Recognition with Recurrent Neural Network model, predict in real time
A possibility that user is likely to purchase product in next step out, i.e. purchase probability.
2. a kind of user's behavior prediction method based on deep learning as described in claim 1, it is characterised in that: the step
In 1, go to identify that different users, user enter a series of clicks of website using the user identity cookie that web log file records
Behavior is arranged according to chronological order, forms behavior access sequence data, the access sequence number for obtaining purchase and not buying
According to sample, in the step 2, the importance for needing to reject includes the buying behavior page lower than desired Web page module, that is, is paid
The necessary page before success, such as information page of insuring, the validation of information page of insuring, the payment page are dosed, it further include entire
In Website page structure, the page being seldom accessed by the user, the elimination method of use is in statistics the page nearest one month by institute
The number for having user to access, the page by access times less than or equal to 3 remove.
3. a kind of user's behavior prediction method based on deep learning as claimed in claim 2, it is characterised in that: the step
Further include following steps in two:
3) data repeatedly bought further are analyzed, a plurality of purchase is intercepted out to buy movement as node to the data repeatedly bought
The access sequence data of movement；
4) length that maximum access sequence is truncated is determined, firstly, the access sequence number do not bought of the filtered access path less than 5 steps
According to, retain the access sequence data of effective access path at least 8 steps, less than 8 steps use space polishing, choose purchase movement forward
Maximum length of 10 steps as access sequence is pushed away, the access sequence greater than 10 is truncated.
4. a kind of user's behavior prediction method based on deep learning as claimed in claim 3, it is characterised in that: the step
In three, Recognition with Recurrent Neural Network uses LSTM algorithm model, and using cumulative form calculus state, training data is divided into two classes, point
Wei sample not bought and not buy sample, data volume is respectively 500,000 or more, and especially up to 1,000,000.
5. a kind of user's behavior prediction method based on deep learning as claimed in claim 3, it is characterised in that: the step
It is that user orients push marketing according to different business scene when the purchase probability of user meets the threshold condition of setting in four
Extension programme, the promotion scheme include providing the marketing tools such as discount coupon, manual service, purchase guidance.
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 true CN109460816A (en) | 2019-03-12 |
| CN109460816B 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) |
Cited By (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 |
| CN113435966A (en) * | 2021-06-22 | 2021-09-24 | 布瑞克农业大数据科技集团有限公司 | Product transaction method and system |
| CN115827947A (en) * | 2023-02-03 | 2023-03-21 | 北京匠数科技有限公司 | Method and device for collecting paging website data and electronic equipment |
| WO2023097518A1 (en) * | 2021-11-30 | 2023-06-08 | Siemens Aktiengesellschaft | Interface display method and apparatus of industrial software |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787100A (en) * | 2016-03-18 | 2016-07-20 | 浙江大学 | User session recommendation method based on deep neural network |
| 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 |
| CN106485562A (en) * | 2015-09-01 | 2017-03-08 | 苏宁云商集团股份有限公司 | A kind of commodity information recommendation method based on user's history behavior and system |
| WO2017093953A1 (en) * | 2015-12-02 | 2017-06-08 | Tata Consultancy Services Limited | Method and system for purchase behavior prediction of customers |
| CN107516235A (en) * | 2016-06-17 | 2017-12-26 | 阿里巴巴集团控股有限公司 | Commodity preference predictor method and device |
| 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 |
| US20180046920A1 (en) * | 2016-08-10 | 2018-02-15 | Paypal, Inc. | User Data Learning Based on Recurrent Neural Networks with Long Short Term Memory |
| 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 |
| US20180293497A1 (en) * | 2017-04-06 | 2018-10-11 | The Boston Consulting Group, Inc. | Systems and Methods for Learning and Predicting Transactions |
| CN108648020A (en) * | 2018-05-15 | 2018-10-12 | 携程旅游信息技术（上海）有限公司 | User behavior quantification method, system, device and storage medium |
| CN108648049A (en) * | 2018-05-03 | 2018-10-12 | 中国科学技术大学 | A kind of sequence of recommendation method based on user behavior difference modeling |
| 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 |
-
2018
- 2018-11-16 CN CN201811365272.6A patent/CN109460816B/en active Active
Patent Citations (13)
| 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 |
| CN106485562A (en) * | 2015-09-01 | 2017-03-08 | 苏宁云商集团股份有限公司 | A kind of commodity information recommendation method based on user's history behavior and system |
| 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 |
| 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 |
| CN107516235A (en) * | 2016-06-17 | 2017-12-26 | 阿里巴巴集团控股有限公司 | Commodity preference predictor 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 |
| US20180293497A1 (en) * | 2017-04-06 | 2018-10-11 | The Boston Consulting Group, Inc. | Systems and Methods for Learning and 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 |
| CN108648049A (en) * | 2018-05-03 | 2018-10-12 | 中国科学技术大学 | A kind of sequence of 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 quantification method, system, device and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| SEBASTIAN HEINZ 等: "An LSTM-based dynamic customer model for fashion recommendation", 《ARXIV:1708.07347》 * |
| 张备: "基于多神经网络的混合动态推荐研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王屯屯: "基于LSTM的在线商城商品推荐研究", 《现代计算机》 * |
| 黄立威 等: "基于深度学习的推荐系统研究综述", 《计算机学报》 * |
Cited By (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 |
| CN113435966A (en) * | 2021-06-22 | 2021-09-24 | 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 |
| CN115827947A (en) * | 2023-02-03 | 2023-03-21 | 北京匠数科技有限公司 | Method and device for collecting paging website data and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109460816B (en) | 2020-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460816A (en) | A kind of user's behavior prediction method based on deep learning | |
| CN104077306B (en) | The result ordering method and system of a kind of search engine | |
| CN107689008A (en) | A kind of user insures the method and device of behavior prediction | |
| CN102956009A (en) | Electronic commerce information recommending method and electronic commerce information recommending device on basis of user behaviors | |
| CN111242318B (en) | Service model training method and device based on heterogeneous feature library | |
| CN101493913A (en) | Method and system for assessing user credit in internet | |
| CN103678659A (en) | E-commerce website cheat user identification method and system based on random forest algorithm | |
| Jain et al. | NFT appraisal prediction: utilizing search trends, public market data, linear regression and recurrent neural networks | |
| CN116362823A (en) | Recommendation model training method, recommendation method and recommendation device for behavior sparse scene | |
| CN113792220A (en) | A target object recommendation method, device, computer equipment and storage medium | |
| CN115880077A (en) | Recommendation method and device based on client label, electronic device and storage medium | |
| CN118446749A (en) | Advertisement prediction optimization method, device and computer equipment based on machine learning | |
| CN108805580A (en) | Account number analysis method, device and storage medium | |
| Pradana et al. | Sentiment analysis of Nanovest investment application using naive bayes algorithm | |
| Kumar Tripathi et al. | Fake review detection in big data using parallel bbo | |
| CN107527289B (en) | Investment portfolio industry configuration method, device, server and storage medium | |
| CN107515928A (en) | A kind of method, apparatus, server, storage medium for judging assets price tendency | |
| CN107622408A (en) | Purchase the predictor method and estimating device of fare rope transaction conversion | |
| CN104834976A (en) | Method for searching for, analyzing and predicting price change trend of memory chip through big data | |
| CN108984668A (en) | A kind of method, apparatus of data processing, equipment and storage medium | |
| CN120278820A (en) | Illegal account identification method and device | |
| Sabri et al. | Click analysis: How e-commerce companies benefit from exploratory and association rule mining | |
| CN108763258A (en) | Document subject matter parameter extracting method, Products Show method, equipment and storage medium | |
| CN110189016A (en) | Technology life cycle appraisal procedure and device | |
| CN112115258A (en) | User credit evaluation method, device, server 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 |