CN110189197A - Electric business personalized recommendation method based on context multi-arm fruit machine - Google Patents

Electric business personalized recommendation method based on context multi-arm fruit machine Download PDF

Info

Publication number
CN110189197A
CN110189197A CN201910428730.4A CN201910428730A CN110189197A CN 110189197 A CN110189197 A CN 110189197A CN 201910428730 A CN201910428730 A CN 201910428730A CN 110189197 A CN110189197 A CN 110189197A
Authority
CN
China
Prior art keywords
information
motion estimation
estimation value
context
recommended
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
CN201910428730.4A
Other languages
Chinese (zh)
Inventor
钟珊
杨馨悦
伏玉琛
应文豪
卫梦
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.)
Changshu Institute of Technology
Original Assignee
Changshu Institute of 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 Changshu Institute of Technology filed Critical Changshu Institute of Technology
Priority to CN201910428730.4A priority Critical patent/CN110189197A/en
Publication of CN110189197A publication Critical patent/CN110189197A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G06F16/355—Class or cluster creation or modification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
    • 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 kind of electric business personalized recommendation methods based on context multi-arm fruit machine, pass through the similarity of computational context information and context multi-arm fruit machine motion characteristic, similarity and model motion estimation value are combined into a new motion estimation value, the new maximum movement of motion estimation value is selected to be recommended.The part that context fruit machine model utilizes is the selection maximum movement of similarity, and contextual information is recommended；The part of exploration is the selection maximum movement of estimated value, corresponding to the hot recommendation in recommender system, finally according to the feedback of user more new model.The present invention carries out movement selection using contextual information and motion estimation value simultaneously, maximizes and awards immediately by utilizing；The information that the potential preference of user is excavated by exploring, increases the diversity of recommendation.Simultaneously when contextual information is unknown, is recommended by exploring, efficiently solve the problems, such as the cold start-up in recommendation field.

Description

Electric business personalized recommendation method based on context multi-arm fruit machine
Technical field
The present invention relates to a kind of electric business personalized recommendation method, more particularly to a kind of based on context multi-arm fruit machine Electric business personalized recommendation method.
Background technique
Intensified learning is the study that intelligent body is mapped from ambient condition to behavior, for solving the problems, such as Sequence Decision.It is any Decision problem is directed to explore and utilize, wherein exploration is to taste using being to select optimal policy according to currently known knowledge experience Try other secondary dominant strategies.It is awarded immediately using maximum can be obtained, but when learning insufficient, algorithm can fall into local optimum, and The award that can sufficiently learn each strategy is explored, optimal policy is found and is not easy that intelligent body is made to fall into local optimum, facilitate maximum Change to accumulate and return, but explore and need to spend more learning times, while slowing down convergence speed of the algorithm, explore and utilizes in fact Border is conflicting.Multi-arm fruit machine (Multi-armed Bandit, referred to as MAB) problem is to balance to explore in intensified learning With the classical problem utilized, single step learning tasks in intensified learning are corresponded to.
One variant of multi-arm fruit machine is context fruit machine (Contextual MAB, referred to as CMAB), is introduced Following traits.The award of CMAB movement meets independently same point with the award for codetermining, and acting is acted by contextual feature Cloth.Current three context fruit machine models most outstanding are: (1) award meets Li Puxici continuity with contextual feature (2) award meets a kind of linear relationship (3) award return one fixed policy class of satisfaction with contextual feature.Online content is recommended It can be modeled as CMAB model.In problems, process recommended to the user corresponds to the movement selection course of CMAB.
Personalized recommendation system recommends article to user according to user's current environment and interest preference.Environmental characteristic and User interest preference is known as contextual information in CMAB model, and contextual information recommends the letter for meeting user interest preference Breath, can obtain the positive feedback of user.But the only information of recommended user's preference can reduce user to the level of interest of information, push away Recommend result while will be rich in novelty.Content-based recommendation algorithm, the algorithm based on collaborative filtering and mixing proposed algorithm are Three kinds of traditional proposed algorithms predict scoring of the user to information by maintenance consumer articles rating matrix, then are pushed away It recommends.There are two main problems for traditional recommender system: (1) being cold-started problem, new user does not have rating matrix, can not be recommended (2) diversity of recommendation information, traditional proposed algorithm is difficult to explore the potential interest preference of user, according only to historical information, Recommendation results do not have novelty.
Summary of the invention
The object of the present invention is to provide a kind of electric business personalized recommendation methods based on context multi-arm fruit machine, sufficiently benefit Recommended with contextual information, while being explored from the higher movement of estimated value, the potential interest preference of Lai Faxian user, Increase the diversity of recommendation information.
The technical scheme is that such: a kind of electric business personalized recommendation side based on context multi-arm fruit machine Method, comprising the following steps:
S1, input data；
S2, the set of actions A and motion characteristic set B for initializing context multi-arm fruit machine model, the set of actions For information aggregate to be recommended, the motion characteristic collection is combined into information characteristics set to be recommended；
S3, the clicking rate that setting context multi-arm fruit machine model motion estimation value Q (i) is information i to be recommended, movement choosing The recommendation number that number T (i)=0 is information i to be recommended and the click volume that accumulation return Sum=0 is information i to be recommended are selected, Middle i ∈ A；
S4, the clicking rate Q (i) for obtaining current all information to be recommended；
S5, judge t moment with the presence or absence of contextual information xt, it is then transferred to step S6 if it exists, is otherwise transferred to step S9, institute State contextual information xtFor the user interest preference feature for being recommended user；
S6, computational context information xtWith everything feature BiSimilarity sim (i), wherein i ∈ A；
S7, according to similarity sim (i) and motion estimation value Q (i), wherein i ∈ A, calculates new motion estimation value Q '；
S8, according to new motion estimation value Q ' carry out recommendation information selection, t moment selection acts k=argmaxiQ ' (i), I ∈ A, goes to step S10；
S9, movement selection is carried out according to motion estimation value Q, t moment selection acts k=argmaxiQ (i), i ∈ A；
S10, by t moment user feedback rt, update accumulation return Sum=Sum+rt, k is acted by selection number Tk=Tk+1 And the estimated value Q (k) of movement k.
Preferably, the step S6 uses cosine similarity computational context information xtWith everything feature BiSimilarity, Make full use of contextual information.
Preferably, similarity sim (i) is obtained into new move multiplied by motion estimation value Q (i) as weight in the step S7 Make estimated value Q ' (i), new motion estimation value calculation formula: Q ' (i)=Q (i) × sim (i), i ∈ A.
Preferably, user feedback r in the step S10tBernoulli Jacob's distribution is obeyed, positive feedback then r is obtainedt=1, it is born Feed back then rt=0.
Preferably, the motion estimation value in the step S13 uses incrementally updating, need to only save current action estimated value With the award of acquisition.Motion estimation value more new formula are as follows: Qt(k)=Qt-1(k)+(rt-Qt-1(k))/t。
Preferably, motion estimation value is set as 1, i.e. Q (i)=1, the clicking rate of all information to be recommended in the step S3 It is equal.
The beneficial effect of technical solution provided by the present invention is,
Context multi-arm fruit machine model can be maximized and be encouraged immediately using being partially that contextual information is acted Reward, exploring part is the selection biggish movement of motion estimation value.Recommender system is modeled as context multi-arm fruit machine model, from The highest information to be recommended of clicking rate, which is set out, to be explored, and is not easy to obtain negative-feedback, helps to maximize accumulation return, discovery is used The information of the potential preference in family, is not easy to obtain Negative Feedback, and the accuracy and diversity for recommending article can be effectively ensured.
Contextual information is made full use of by cosine similarity, contextual feature and motion characteristic similarity are calculated, by phase It is combined into a new motion estimation value with motion estimation value like degree, movement selection is carried out according to new motion estimation value, upper Hereafter contextual information is taken full advantage of in multi-arm fruit machine model.
Detailed description of the invention
Fig. 1 is the method for the present invention flow diagram；
Fig. 2 is Yahoo used in the embodiment of the present invention！R6A recommending data collection；
Fig. 3 is Yahoo used in the embodiment of the present invention！The data line that R6A recommending data is concentrated；
Fig. 4 is that the embodiment of the present invention is based on Yahoo！Recommender system is modeled as context multi-arm fruit machine by R6A data set The algorithm frame figure of model；
Fig. 5 is the embodiment of the present invention in Yahoo！Recommendation effect schematic diagram on R6A data set；
Fig. 6 is the embodiment of the present invention in Yahoo！Recommend to tie with other context multi-arm fruit machine algorithms on R6A data set Fruit comparison schematic diagram.
Specific embodiment
Below with reference to embodiment, the invention will be further described, but not as a limitation of the invention.
Incorporated by reference to shown in Fig. 1, the present invention is based on the electric business personalized recommendation method of context multi-arm fruit machine, including it is following Step:
S1, input Yahoo！R6A data set；
S2, the set of actions A and motion characteristic set B for initializing context multi-arm fruit machine model, the set of actions For information aggregate to be recommended, the motion characteristic collection is combined into information characteristics set to be recommended；
S3, it the clicking rate that setting context multi-arm fruit machine model motion estimation value Q (i)=1 is information i to be recommended, moves Elect the recommendation number that number T (i)=0 is information i to be recommended and the click that accumulation return Sum=0 is information i to be recommended Amount, wherein i ∈ A, the clicking rate of all information to be recommended are equal；
S4, the clicking rate Q (i) for obtaining current all information to be recommended；
S5, judge t moment with the presence or absence of contextual information xt, it is then transferred to step S6 if it exists, is otherwise transferred to step S9, institute State contextual information xtFor the user interest preference feature for being recommended user；
S6, using cosine similarity computational context information xtWith everything feature BiSimilarity sim (i) makes full use of Contextual information, wherein i ∈ A；
S7, according to similarity sim (i) and motion estimation value Q (i), wherein i ∈ A, similarity sim (i) is multiplied as weight New motion estimation value Q ', new motion estimation value calculation formula: Q ' (i)=Q (i) × sim are calculated with motion estimation value Q (i) (i), i ∈ A；
S8, according to new motion estimation value Q ' carry out recommendation information selection, t moment selection acts k=argmaxiQ ' (i), I ∈ A, goes to step S10；A possibility that biggish movement of estimated value Q and the selected biggish movement of similarity sim, is all larger. The biggish movement of sim is the part that context multi-arm fruit machine model utilizes, and based on context carries out movement selection, can be maximum Change is awarded immediately；The biggish movement of Q is the part that context multi-arm fruit machine model is explored, from the higher movement of estimated value It is explored, attempts other movements of selection, be not easy to obtain the negative-feedback of environment, help to maximize accumulation return.
S9, movement selection is carried out according to motion estimation value Q, t moment selection acts k=argmaxiQ (i), i ∈ A；It will push away Recommending system modelling is context multi-arm fruit machine model, recommends the highest movement of clicking rate to belong to hot recommendation, meets most use The interest preference at family, when lacking contextual information, hot recommendation is not easy to obtain the negative-feedback of user.
S10, by t moment user feedback rt, update accumulation return Sum=Sum+rt, k is acted by selection number Tk=Tk+1 And the estimated value Q (k) of movement k need to only save the award of current action estimated value and acquisition, movement using incrementally updating Estimated value more new formula are as follows: Qt(k)=Qt-1(k)+(rt-Qt-1(k))/t.User feedback rtBernoulli Jacob's distribution is obeyed, is obtained positive and negative Present then rt=1, obtain negative-feedback then rt=0.
It is shown in Figure 2, Yahoo！R6A data set meets the End Data Protection standard of Yahoo, by audit dedicated for pushing away Recommend system.R6A data set includes Yahoo！Point of the news article that Today module is shown in preceding ten days users in May, 2009 Hit log.The data set includes 45,811,883 users to the access log of Today module, and user and every article all use 6 Dimensional vector indicates its feature.Contextual information in Yahoo's R6A data set is exactly the feature vector of 6 dimensions.Wherein, access user is special Sign includes: gender, age, regional information, historical behavior classification；Article feature includes: that article comes origin url, article category (institute Belong to label).User characteristics vector is normalized to unit vector first, then carries out dimensionality reduction, the user characteristics of 1,000 multidimensional are reflected It is mapped to 80 multidimensional article feature spaces.It is clustered respectively according to the user characteristics after article feature and dimensionality reduction, obtains 5 texts Chapter class cluster and 5 user class clusters finally add constant characteristic 1, indicate user and article feature with 6 dimensional vectors.Existing research card The effect of bright selected 6 dimensional features is optimal, while also reducing computation complexity and memory space.The context that the present invention discusses is special Reference breath is exactly the feature vector of 6 dimensions in Yahoo's media recommender data set.
It is shown in Figure 3, it is Yahoo shown in Fig. 2！Data line in R6A data set.Every row respectively indicates in Fig. 3: Whether the timestamp of user's access, article ID, the user of actual displayed click mark, and (1 indicates that user clicks, 0 indicates non-point Hit), user characteristics, all articles on the same day and the feature of article.
It is 0 that whether most record users, which click mark, in data set, that is, when showing this article, does not obtain the anti-of user Feedback, these records can not be used as experimental data.Only user, which clicks article, just indicates that user is interested in this article.It is first First, these invalid datas are filtered out, picks out the record of all users' clicks as experimental data.
With data instance on May 1st, 2009, obtain following result after crossing filter data: the same day shows 49 articles altogether, uses The click information at family records for 2107 totally, and wherein user at least clicks one, at most clicks 49.It will click on more than two The user of article, the target user as recommendation.Wherein article going through as user is selected from the article that user clicks Records of the Historian record, this article feature is user characteristics, and the article that other are clicked is as test data.First when recommending for user User characteristics are obtained, user characteristics and the same day all articles are calculated into characteristic similarity, using similarity as weight multiplied by movement Estimated value is recommended according to new estimated value to user as new motion estimation value, and test data is used to judge that user is The no article that can click recommendation.
It is shown in Figure 4, according to Yahoo！News recommender system is modeled as context multi-arm fruit machine by R6A data set The algorithm frame of model.Recommendation process corresponds to context multi-arm fruit machine model and acts selection course.Wherein, news article pair Context multi-arm fruit machine model should be corresponded in the movement of context multi-arm fruit machine model, article feature and user characteristics Contextual information.Motion estimation value corresponds to the clicking rate of news article, movement selection time in context multi-arm fruit machine model Number corresponds to news article and is recommended number, and accumulation return corresponds to the accumulation number of clicks of user.User characteristics are that user goes through The article feature that history is clicked.
When being recommended, the user characteristics of recommendation are first obtained, calculate user characteristics and all article features to be recommended Characteristic similarity sim is combined into a new movement with article motion estimation value Q to be recommended as weight and estimated by similarity sim Evaluation Q '=sim × Q recommends to select the new maximum article of estimated value Q ' every time.By calculating user characteristics and needing to be pushed away The similarity for recommending article feature recommends to meet the article of its interest preference to user, at the same from the high article of clicking rate into Row is explored, and attempts to recommend other articles, the potential interested article of Lai Faxian user increases the range of recommendation.Observation pushes away The feedback for recommending rear user updates article clicking rate according to user feedback, article is recommended number and user accumulates click volume.
It is shown in Figure 5, in Yahoo！Emulation experiment is recommended on R6A data set.Choose Yahoo！In R6A data set On May 1st, 2009, data, recommended for 2107 users, and every user recommends 10 articles, and user, which accumulates, after recommendation clicks Amount is as shown in Figure 5.
It is shown in Figure 6, in Yahoo！Emulation experiment is recommended on R6A data set.Context algorithm, according only to upper Context information is recommended, and recommends and the most like article of user characteristics；Clicks algorithm is pushed away according only to article clicking rate It recommends, recommends the highest article of clicking rate；RandomChoice algorithm, random selection article are recommended；LinUCB algorithm is a kind of warp The context fruit machine algorithm of allusion quotation, it is assumed that user feedback and article feature are in a linear relationship, and the news for being applied to Yahoo is recommended In system.Choose Yahoo！Data on May 1st, 2009 in R6A data set are recommended for 2107 users, and every user pushes away 10 articles are recommended, compared with above-mentioned four kinds of context proposed algorithms, the present invention is based on the recommendation of context multi-arm fruit machine calculations Method effect is optimal, and recommendation results are as shown in Figure 6.

Claims (6)

1. a kind of electric business personalized recommendation method based on context multi-arm fruit machine, which comprises the following steps:
S1, input data；
S2, initialize context multi-arm fruit machine model set of actions A and motion characteristic set B, the set of actions be to Recommendation information set, the motion characteristic collection are combined into information characteristics set to be recommended；
S3, the clicking rate that setting context multi-arm fruit machine model motion estimation value Q (i) is information i to be recommended, movement selection time The recommendation number and accumulation that number T (i)=0 is information i to be recommended return the click volume that Sum=0 is information i to be recommended, wherein i ∈ A；
S4, the clicking rate Q (i) for obtaining current all information to be recommended；
S5, judge t moment with the presence or absence of contextual information xt, be then transferred to step S6 if it exists, be otherwise transferred to step S9, it is described on Context information xtFor the user interest preference feature for being recommended user；
S6, computational context information xtWith everything feature BiSimilarity sim (i), wherein i ∈ A；
S7, according to similarity sim (i) and motion estimation value Q (i), wherein i ∈ A, calculates new motion estimation value Q '；
S8, according to new motion estimation value Q ' carry out recommendation information selection, t moment selection acts k=argmaxiQ ' (i), i ∈ A, Go to step S10；
S9, movement selection is carried out according to motion estimation value Q, t moment selection acts k=argmaxiQ (i), i ∈ A；
S10, by t moment user feedback rt, update accumulation return Sum=Sum+rt, k is acted by selection number Tk=Tk+ 1 and dynamic Make the estimated value Q (k) of k.
2. the electric business personalized recommendation method according to claim 1 based on context multi-arm fruit machine, which is characterized in that The step S6 uses cosine similarity computational context information xtWith everything feature BiSimilarity.
3. the electric business personalized recommendation method according to claim 1 based on context multi-arm fruit machine, which is characterized in that Similarity sim (i) is obtained into new motion estimation value Q ' (i) multiplied by motion estimation value Q (i) as weight in the step S7, New motion estimation value calculation formula: Q ' (i)=Q (i) × sim (i), i ∈ A.
4. the electric business personalized recommendation method according to claim 1 based on context multi-arm fruit machine, which is characterized in that User feedback r in the step S10tBernoulli Jacob's distribution is obeyed, positive feedback then r is obtainedt=1, obtain negative-feedback then rt=0.
5. the electric business personalized recommendation method according to claim 1 based on context multi-arm fruit machine, which is characterized in that Motion estimation value in the step S13 uses incrementally updating, motion estimation value more new formula are as follows: Qt(k)=Qt-1(k)+ (rt-Qt-1(k))/t。
6. the electric business personalized recommendation method according to claim 1 based on context multi-arm fruit machine, which is characterized in that Motion estimation value is set as 1, i.e. Q (i)=1 in the step S3, and the clicking rate of all information to be recommended is equal.
CN201910428730.4A 2019-05-22 2019-05-22 Electric business personalized recommendation method based on context multi-arm fruit machine Pending CN110189197A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910428730.4A CN110189197A (en) 2019-05-22 2019-05-22 Electric business personalized recommendation method based on context multi-arm fruit machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910428730.4A CN110189197A (en) 2019-05-22 2019-05-22 Electric business personalized recommendation method based on context multi-arm fruit machine

Publications (1)

Publication Number Publication Date
CN110189197A true CN110189197A (en) 2019-08-30

Family

ID=67717366

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910428730.4A Pending CN110189197A (en) 2019-05-22 2019-05-22 Electric business personalized recommendation method based on context multi-arm fruit machine

Country Status (1)

Country Link
CN (1) CN110189197A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112256739A (en) * 2020-11-12 2021-01-22 同济大学 Method for screening data items in dynamic flow big data based on multi-arm gambling machine
CN112733004A (en) * 2021-01-22 2021-04-30 上海交通大学 Movie and television work recommendation method based on multi-arm tiger machine algorithm

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2202656A1 (en) * 2008-12-23 2010-06-30 Axel Springer Digital TV Guide GmbH Context-based recommender system
CN103297853A (en) * 2013-06-07 2013-09-11 华东师范大学 IPTV (internet protocol television) program recommendation method based on context recognition for multiple users
CN106886389A (en) * 2017-03-22 2017-06-23 苏州大学 The random multi-arm decision problem computational methods and its device of a kind of self adaptation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2202656A1 (en) * 2008-12-23 2010-06-30 Axel Springer Digital TV Guide GmbH Context-based recommender system
CN103297853A (en) * 2013-06-07 2013-09-11 华东师范大学 IPTV (internet protocol television) program recommendation method based on context recognition for multiple users
CN106886389A (en) * 2017-03-22 2017-06-23 苏州大学 The random multi-arm decision problem computational methods and its device of a kind of self adaptation

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
周倩: "大规模多臂赌博机算法研究与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
张佃磊: "基于自适应上下文多臂赌博机推荐算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112256739A (en) * 2020-11-12 2021-01-22 同济大学 Method for screening data items in dynamic flow big data based on multi-arm gambling machine
CN112256739B (en) * 2020-11-12 2022-11-18 同济大学 Method for screening data items in dynamic flow big data based on multi-arm gambling machine
CN112733004A (en) * 2021-01-22 2021-04-30 上海交通大学 Movie and television work recommendation method based on multi-arm tiger machine algorithm

Similar Documents

Publication Publication Date Title
CN110110181A (en) A kind of garment coordination recommended method based on user styles and scene preference
CN101751437A (en) Web active retrieval system based on reinforcement learning
CN103514239A (en) Recommendation method and system integrating user behaviors and object content
WO2021139415A1 (en) Data processing method and apparatus, computer readable storage medium, and electronic device
CN109711925A (en) Cross-domain recommending data processing method, cross-domain recommender system with multiple auxiliary domains
CN102968506A (en) Personalized collaborative filtering recommendation method based on extension characteristic vectors
CN107545471B (en) Big data intelligent recommendation method based on Gaussian mixture
CN106897914A (en) A kind of Method of Commodity Recommendation and system based on topic model
Li et al. Content-based filtering recommendation algorithm using HMM
Wu et al. A hybrid-forecasting model reducing Gaussian noise based on the Gaussian support vector regression machine and chaotic particle swarm optimization
CN104239496A (en) Collaborative filtering method based on integration of fuzzy weight similarity measurement and clustering
CN112700274A (en) Advertisement click rate estimation method based on user preference
Alshmrany Adaptive learning style prediction in e-learning environment using levy flight distribution based CNN model
CN109903138A (en) A kind of individual commodity recommendation method
CN114969533A (en) Sequence recommendation method based on long-term and short-term preference of user
CN110189197A (en) Electric business personalized recommendation method based on context multi-arm fruit machine
Lang et al. Movie recommendation system for educational purposes based on field-aware factorization machine
Wang et al. Attention-based deep neural network for internet platform group users’ dynamic identification and recommendation
CN114896515A (en) Time interval-based self-supervision learning collaborative sequence recommendation method, equipment and medium
CN103136309A (en) Method for carrying out modeling on social intensity through learning based on core
CN113221003B (en) Mixed filtering recommendation method and system based on dual theory
CN108984551A (en) A kind of recommended method and system based on the multi-class soft cluster of joint
Pang et al. Efficient deep reinforcement learning-enabled recommendation
CN113688306A (en) Recommendation strategy generation method and device based on reinforcement learning
CN116975686A (en) Method for training student model, behavior prediction method and device

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20190830