CN112612948B - Deep reinforcement learning-based recommendation system construction method - Google Patents

Deep reinforcement learning-based recommendation system construction method Download PDF

Info

Publication number
CN112612948B
CN112612948B CN202011473950.8A CN202011473950A CN112612948B CN 112612948 B CN112612948 B CN 112612948B CN 202011473950 A CN202011473950 A CN 202011473950A CN 112612948 B CN112612948 B CN 112612948B
Authority
CN
China
Prior art keywords
user
recommendation system
module
actor
items
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
Application number
CN202011473950.8A
Other languages
Chinese (zh)
Other versions
CN112612948A (en
Inventor
石龙翔
金苍宏
李卓蓉
吴明晖
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.)
Hangzhou Woerfu Information Technology Co ltd
Original Assignee
Hangzhou City University
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 Hangzhou City University filed Critical Hangzhou City University
Priority to CN202011473950.8A priority Critical patent/CN112612948B/en
Publication of CN112612948A publication Critical patent/CN112612948A/en
Application granted granted Critical
Publication of CN112612948B publication Critical patent/CN112612948B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—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
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Human Resources & Organizations (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Educational Administration (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention provides a recommendation system construction method based on deep reinforcement learning, which comprises the following steps: s1) establishing a feature characterization set of the interaction between the user and the recommendation system; s2) establishing a state representation of the interaction between the user and the recommendation system; s3) modeling of a recommendation system; s4) training of a recommendation system; s5) recommending the deployment of the system. The invention has the advantages that: the action selection space of the one-dimensional discrete type project is mapped to the multi-dimensional continuous real-valued space, and the action space of the recommended item is simplified by adopting a binary conversion method, so that the training difficulty of a recommendation system is reduced; the behavior characteristics of the user are modeled by adopting the convolution recurrent neural network, so that the performance of the recommendation system is improved.

Description

Deep reinforcement learning-based recommendation system construction method
Technical Field
The invention relates to the technical field of recommendation systems, in particular to a recommendation system construction method based on deep reinforcement learning.
Background
The recommendation system is an indispensable part of an intelligent electronic commerce system, and the recommendation of possibly required items to a user through historical browsing data of the user is a main task of the recommendation system. Common recommendation system methods are collaborative filtering, matrix factorization, content-based ranking, and the like. However, these conventional recommendation methods often model the user preference as a static process, and recommend to the user by some greedy ranking methods, which cannot consider the dynamic change of the user preference. Recent studies show that modeling of the recommendation system is a reinforcement learning problem, and by maximizing total scores which may be given by a user in the future as an optimization target, dynamic modeling can be effectively performed on the preference of the user, and the performance and performance of the recommendation system are improved.
However, due to some characteristics of the recommendation system itself, the following two problems are often encountered in the process of constructing the recommendation system by using the reinforcement learning method: 1) the number of recommended items in a recommendation system is often huge, and optimization difficulty caused by the huge action space problem needs to be considered when the recommendation system is used for solving the problem by reinforcement learning; 2) the modeling of the observed state of the user can generate time sequence information fed back by the user in the interaction process of the user and the recommendation system, and how to extract the time sequence information, model the state of the user and describe the behavior characteristics of the user is a key problem about the performance of the recommendation system.
Disclosure of Invention
The invention aims to provide a recommendation system construction method based on deep reinforcement learning, which effectively simplifies an action space by mapping the action selection space of a one-dimensional discrete item to a multi-dimensional continuous real-valued space and improves the performance of a recommendation system by adopting a convolution recurrent neural network to model the action characteristics of a user.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a recommendation system construction method based on deep reinforcement learning comprises the following steps:
s1) establishing a feature characterization set of user interaction with a recommendation system
The interaction process of the user and the recommendation system adopts a Markov decision process<S,A,P,R>And representing, wherein S is a state set, namely a characteristic characterization set of user interaction with the recommendation system, A is a set of items selectable by the user, and the items selected at the time t are marked as at，P(st+1|st,at) As a function of the state transition, i.e. the state s at the present momenttNext user selects item atLast next moment state st+1Probability of (a), R(s)t+1|st,at) As a function of reward, i.e. user vs. current status stSelecting item atRating of the rear user, the optimization goal of the recommendation system being to maximize the expected total rating of the user, i.e.
Figure BDA0002834456300000021
Wherein T is the termination time;
s2) establishing state representation of interaction between user and recommendation system
Establishing a user scoring matrix with U rows and M columns according to the historical records of all users, wherein U is the total number of the users, M is the total number of recommended items, the ith row and the jth column of the scoring matrix represent the scoring of the ith user on the jth item, if the user does not evaluate the item, the scoring is set to be 0, and each column of the matrix is used as the characteristic of each recommended item;
extracting all the selected items and scores before the current moment according to the historical records (including items selected by the user and the scores of the user) in the interaction process of each user and the recommendation system, inputting the items and the scores into a convolutional neural network, processing the output of all LSTM (time-circulating neural network) units by using a convolutional layer, and processing the output of the convolutional layer by using maximum pooling and mean pooling to serve as the state representation of the interaction of the user and the recommendation system;
wherein, the neural network parameters representing the user state can be learned through the training process of the step S4);
s3) modeling of recommendation system
Modeling a recommendation system by adopting an Actor-critic (Actor-critic) framework, wherein an Actor module inputs a currently observed user state and outputs a recommended item given based on the user state, and a critic module receives the current state and a representation of a selected item as input, is responsible for evaluating selectable items under the current user observed state and evaluating expected returns which may be received in the future;
the actor module and the critics module are both expressed by adopting a neural network;
in order to process the large number of recommended items, project codes for mapping index codes of the recommended items to continuous real values in a B system are adopted, so that the action output dimension of an actor module is logBM of each output valueIn the range of [0, B-1]；
S4) training of recommendation system
Training an actor network and a critic network by using historical data of a user by using a deterministic strategy gradient algorithm (DDPG);
specifically, the scoring data of each user in the training data is first processed into<st,at,rt,st+1>In quadruple form of (a), wherein stIs the state of the user at time t, atB-ary coding, r, for selecting an entry for a usertAs a score of the user, st+1The state at the time t + 1; then, a deterministic strategy gradient algorithm DDPG is adopted, user data in a four-tuple form is stored in experience cache of the DDPG, and an actor network and a critic network are trained;
s5) deployment of recommendation system
On an online deployment platform, recommending corresponding projects by utilizing the actor network trained in the step S4) according to the current state of the user by combining the output values of the actor module and the critic module;
specifically, on the online deployment platform, using the actor module trained in step S4), the actor module outputs an item code according to the current state of the user, matching the output item code with all item codes using a similarity evaluation method, selecting K items with the largest similarity, evaluating the output value of the critics module in the current state, and selecting N items with the largest output value of the critics module as recommended items to be output to the user.
Compared with the prior art, the invention has the following advantages:
according to the recommendation system construction method based on deep reinforcement learning, action selection spaces of one-dimensional discrete items are mapped to a multi-dimensional continuous real-valued space, and an action space of recommended items is simplified by adopting a method of binary conversion, so that the difficulty of training of a recommendation system is reduced; the behavior characteristics of the user are modeled by adopting the convolution recurrent neural network, so that the performance of the recommendation system is improved.
Drawings
FIG. 1 is a schematic diagram illustrating a state representation of user interaction with a recommendation system in an embodiment of the present invention.
Fig. 2 is a schematic diagram of an actor module modeled by the recommendation system in an embodiment of the invention.
FIG. 3 is a schematic diagram of a critics module modeling the recommendation system in an embodiment of the invention.
Detailed Description
Embodiments of the present invention are described in further detail below with reference to the accompanying drawings.
A recommendation system construction method based on deep reinforcement learning comprises the following steps:
s1) establishing a feature characterization set of user interaction with a recommendation system
The interaction process of the user and the recommendation system adopts a Markov decision process<S,A,P,R>And representing, wherein S is a state set, namely a characteristic characterization set of user interaction with the recommendation system, A is a set of items selectable by the user, and the items selected at the time t are marked as at，P(st+1|st,at) As a function of the state transition, i.e. the state s at the present momenttNext user selects item atLast next moment state st+1Probability of (a), R(s)t+1|st,at) As a function of reward, i.e. user vs. current status stSelecting item atRating of the rear user, the optimization goal of the recommendation system is to maximize the total desired rating of the user, i.e. the total rating of the user
Figure BDA0002834456300000041
Wherein T is the termination time.
S2) establishing the state representation of the interaction between the user and the recommendation system
Establishing a user scoring matrix with U rows and M columns according to the historical records of all users, wherein U is the total number of the users, M is the total number of recommended items, the ith row and the jth column of the scoring matrix represent the scoring of the ith user on the jth item, if the user does not evaluate the item, the scoring is set to be 0, and each column of the matrix is used as the characteristic of each recommended item;
extracting all the selected items and scores before the current moment according to the historical records (including items selected by the user and the scores of the user) in the interaction process of each user and the recommendation system, inputting the items and the scores into a convolutional neural network, processing the output of all LSTM (time-circulating neural network) units by using a convolutional layer, and processing the output of the convolutional layer by using maximum pooling and mean pooling to serve as the state representation of the interaction of the user and the recommendation system;
the neural network parameters representing the user state can be learned through the training process of step S4), and the neural network representing the user state can be used as a part of the actor-critic framework, and the representation of the interaction state of a specific user and the recommendation system is shown in fig. 1.
S3) modeling of recommendation system
An Actor-critic (Actor-critic) framework is used to model the recommendation system, wherein the Actor module inputs the currently observed user state and outputs recommended items given based on the user state, and the critic module receives as input the current state and representations of selected items, and is responsible for evaluating the selectable items in the current user observed state and evaluating the expected reward that it may receive in the future, namely:
Figure BDA0002834456300000051
the actor module and the critics module are both expressed by adopting a neural network;
in order to process the large number of recommended items, item codes for mapping index codes of the recommended items to continuous real values in a B system are adopted, so that the action output dimension of an actor module is logBM, each output value of which ranges from [0, B-1 ]]The actor module outputs an index code of the recommended item, and the actor module and the critics module are respectively shown in fig. 2 and 3.
S4) training of recommendation system
Training an actor network and a critic network by using historical data of a user by using a deterministic strategy gradient algorithm (DDPG);
specifically, the scoring data of each user in the training data is first processed into<st,at,rt,st+1>In quadruple form of (a), wherein stIs the state of the user at time t, atB-ary coding, r, for selecting an entry for a usertAs a score of the user, st+1The state at the time t + 1; then, a deterministic strategy gradient algorithm DDPG is adopted, user data in a four-tuple form is stored in experience cache of the DDPG, and an actor network and a critic network are trained;
the detailed training algorithm is described with reference to the following algorithm 1, wherein the hyper-parameters include: a discount factor gamma, a learning time v, a learning sample number w and a soft updating factor tau.
Figure BDA0002834456300000061
S5) deployment of recommendation system
On an online deployment platform, recommending corresponding projects by utilizing the actor network trained in the step S4) according to the current state of the user by combining the output values of the actor module and the critic module;
specifically, on an online deployment platform, using the actor module trained in step S4), the actor module outputs an entry code according to the current state of the user, and using a similarity evaluation method to perform similarity matching between the output entry code and all entry codes, where the similarity matching may select a judgment method of euclidean distance or cosine similarity, select K items with the largest similarity, evaluate the output value of the critic module in the current state, and select N items with the largest output value of the critic module as recommended entries to be output to the user, and the specific recommendation process is shown in the following algorithm 2.
Figure BDA0002834456300000071
Experimental results of the invention experiments were performed on the MovieLens-1M dataset (https:// group. org/dates/MovieLens/1M /) commonly used by recommendation systems, which contained score data for about 6000 users about 4000 movies, for a total of 1,000,000 scores. 80% of the data were taken as training set and 20% as test set.
The experimental effect of the invention was compared with the effects obtained by several other recommendation methods, including popularity-based recommendation and Singular Value composition (SVD). As shown in table 1, by comparing with the two methods, including statistics of precision (precision) and recall (recall) of 30 items, a higher value indicates better recommendation effect, so that the method of the present invention is very effective.
TABLE 1 comparison of the effectiveness of the method with several other types of recommendation methods
Recommendation method Precision ratio Recall rate
Popularity-based recommendations 32.4％ 0.713％
SVD 30.3％ 0.558％
Method for producing a composite material 45.7％ 2.12％
The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the spirit of the present invention, and these modifications and improvements should also be considered as within the scope of the present invention.

Claims (3)

1. A recommendation system construction method based on deep reinforcement learning is characterized by comprising the following steps:
s1) establishing a feature characterization set of the interaction between the user and the recommendation system
The interaction process of the user and the recommendation system adopts a Markov decision process<S,A,P,R>Showing, wherein S is a state set, namely a characteristic characterization set interacted with a recommendation system by a user, A is a set of items selectable by the user, and the items selected at the time t are marked as at，P(st+1|st,at) As a function of the state transition, i.e. the state s at the present momenttNext user selects item atState s at the next momentt+1Probability of (a), R(s)t+1|st,at) As a function of the return, i.e. the user is given the current status stSelecting item atScoring of the later user, wherein the optimization goal of the recommendation system is to maximize the total score expected by the user;
s2) establishing the state representation of the interaction between the user and the recommendation system
Establishing a user scoring matrix with U rows and M columns according to the historical records of all users, wherein U is the total number of the users, M is the total number of recommended items, the ith row and the jth column of the scoring matrix represent the scoring of the ith user on the jth item, if the user does not evaluate the item, the scoring is set to be 0, and each column of the matrix is used as the characteristic of each recommended item;
selecting historical selection items and scores of the users according to historical records in the interaction process of each user and the recommendation system, modeling each historical item by adopting a recurrent neural network, processing the output of the recurrent neural network by adopting the recurrent neural network, and processing the characteristics of the recurrent neural network by adopting a pooling layer to serve as the characteristic representation of the interaction of the user and the recommendation system;
s3) modeling of recommendation system
Modeling a recommendation system by adopting an Actor-critic (Actor-critic) framework, wherein an Actor module inputs a currently observed user state and outputs a recommended item given based on the user state, and a critic module receives the current state and a representation of a selected item as input, is responsible for evaluating selectable items under the state observed by the current user and evaluating expected returns which the critic module may receive in the future;
the actor module and the critics module are both expressed by adopting a neural network; wherein, the actor module adopts a method of system conversion to map items of the recommendation system, and represents the code of each recommendation item by a B system, and the actor module neural network output is logBM real values each ranging from [0, B-1 ]]；
S4) training of recommendation system
Training an actor network and a critic network by using historical data of a user by using a deterministic strategy gradient algorithm (DDPG);
s5) deployment of recommendation system
And on the online deployment platform, recommending corresponding projects by using the actor network trained in the step S4) according to the current state of the user and combining the output values of the actor module and the critic module.
2. The deep reinforcement learning-based recommendation system construction method according to claim 1, wherein:
in the step S4), the method for training the actor module and the critic module using the historical data includes: extracting feature S of the current time from the score data of each user according to the feature representation method of step S2)tAnd the feature s of the next timet+1Current user rating rtAs the instant report, the index number of the item selected by the user is converted into B-system code as the action a, and the action a is stored<st,at,rt,st+1>To DDPG algorithmAnd training the actor module and the critics module by using a DDPG algorithm.
3. The deep reinforcement learning-based recommendation system construction method according to claim 1, wherein:
in step S5), the deployment method of the recommendation system includes: and obtaining codes of the selected items according to the current state input to the actor module, calculating the similarity between the B-system codes of all recommended items and the output codes of the actor module by using a similarity evaluation method, selecting K items with the maximum similarity, evaluating the output value of the critic module in the current state, and selecting N items with the maximum output of the critic module as recommended items to be output to a user.
CN202011473950.8A 2020-12-14 2020-12-14 Deep reinforcement learning-based recommendation system construction method Active CN112612948B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011473950.8A CN112612948B (en) 2020-12-14 2020-12-14 Deep reinforcement learning-based recommendation system construction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011473950.8A CN112612948B (en) 2020-12-14 2020-12-14 Deep reinforcement learning-based recommendation system construction method

Publications (2)

Publication Number Publication Date
CN112612948A CN112612948A (en) 2021-04-06
CN112612948B true CN112612948B (en) 2022-07-08

Family

ID=75234216

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011473950.8A Active CN112612948B (en) 2020-12-14 2020-12-14 Deep reinforcement learning-based recommendation system construction method

Country Status (1)

Country Link
CN (1) CN112612948B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112883285B (en) * 2021-04-28 2021-08-13 北京搜狐新媒体信息技术有限公司 Information recommendation method and device
CN113449183B (en) * 2021-06-18 2022-07-12 华中科技大学 Interactive recommendation method and system based on offline user environment and dynamic rewards
CN114417124A (en) * 2021-11-30 2022-04-29 哈尔滨工程大学 Multi-task reinforcement learning recommendation method
CN116932893B (en) * 2023-06-21 2024-06-04 江苏大学 Sequence recommendation method, system, equipment and medium based on graph convolution network

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109062919A (en) * 2018-05-31 2018-12-21 腾讯科技（深圳）有限公司 A kind of content recommendation method and device based on deeply study
US10229205B1 (en) * 2016-10-14 2019-03-12 Slack Technologies, Inc. Messaging search and management apparatuses, methods and systems
CN110569443A (en) * 2019-03-11 2019-12-13 北京航空航天大学 Self-adaptive learning path planning system based on reinforcement learning
CN111860649A (en) * 2020-07-21 2020-10-30 赵佳 Action set output method and system based on multi-agent reinforcement learning

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059262B (en) * 2019-04-19 2021-07-02 武汉大学 Project recommendation model construction method and device based on hybrid neural network and project recommendation method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10229205B1 (en) * 2016-10-14 2019-03-12 Slack Technologies, Inc. Messaging search and management apparatuses, methods and systems
CN109062919A (en) * 2018-05-31 2018-12-21 腾讯科技（深圳）有限公司 A kind of content recommendation method and device based on deeply study
CN110569443A (en) * 2019-03-11 2019-12-13 北京航空航天大学 Self-adaptive learning path planning system based on reinforcement learning
CN111860649A (en) * 2020-07-21 2020-10-30 赵佳 Action set output method and system based on multi-agent reinforcement learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于深度强化学习的推荐系统研究;刘洋军;《中国优秀博硕士学位论文全文数据库 信息科技辑》;20200115;第20页第3章到第61页第5章 *

Also Published As

Publication number Publication date
CN112612948A (en) 2021-04-06

Similar Documents

Publication Publication Date Title
CN112612948B (en) Deep reinforcement learning-based recommendation system construction method
CN106021364B (en) Foundation, image searching method and the device of picture searching dependency prediction model
CN111538908B (en) Search ranking method and device, computer equipment and storage medium
CN111444394B (en) Method, system and equipment for obtaining relation expression between entities and advertisement recall system
CN111797321B (en) Personalized knowledge recommendation method and system for different scenes
CN103116639B (en) Based on article recommend method and the system of user-article bipartite graph model
CN108573399B (en) Merchant recommendation method and system based on transition probability network
CN104268142B (en) Based on the Meta Search Engine result ordering method for being rejected by strategy
CN110647995A (en) Rule training method, device, equipment and storage medium
CN112131261A (en) Community query method and device based on community network and computer equipment
CN112529415A (en) Article scoring method based on combined multi-receptive-field-map neural network
CN112612951A (en) Unbiased learning sorting method for income improvement
CN109214882A (en) A kind of Method of Commodity Recommendation
CN117635238A (en) Commodity recommendation method, device, equipment and storage medium
CN111079011A (en) Deep learning-based information recommendation method
CN113342994B (en) Recommendation system based on non-sampling cooperative knowledge graph network
CN118014652A (en) Advertisement creative design method and system based on artificial intelligence and big data analysis technology
CN113034193A (en) Working method for modeling of APP2VEC in wind control system
CN115455279A (en) Recommendation method based on deep reinforcement learning recommendation system with high calculation efficiency
He et al. Semisupervised network embedding with differentiable deep quantization
CN112905591B (en) Data table connection sequence selection method based on machine learning
CN115422396A (en) Implicit video recommendation algorithm
CN114880490A (en) Knowledge graph completion method based on graph attention network
CN111967600B (en) Feature derivation method based on genetic algorithm in wind control scene
CN114330291A (en) Text recommendation system based on dual attention mechanism

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
TR01 Transfer of patent right

Effective date of registration: 20240314

Address after: Room B407, Building 2, No. 452, 6th Street, Baiyang Street, Hangzhou Economic and Technological Development Zone, Zhejiang Province, 310018

Patentee after: Hangzhou Woerfu Information Technology Co.,Ltd.

Country or region after: China

Address before: 310000 No.51 Huzhou street, Gongshu District, Hangzhou City, Zhejiang Province

Patentee before: HANGZHOU City University

Country or region before: China

TR01 Transfer of patent right