CN113158024B - Causal reasoning method for correcting popularity deviation of recommendation system - Google Patents

Causal reasoning method for correcting popularity deviation of recommendation system Download PDF

Info

Publication number
CN113158024B
CN113158024B CN202110218946.5A CN202110218946A CN113158024B CN 113158024 B CN113158024 B CN 113158024B CN 202110218946 A CN202110218946 A CN 202110218946A CN 113158024 B CN113158024 B CN 113158024B
Authority
CN
China
Prior art keywords
user
item
recommendation system
scores
popularity
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
CN202110218946.5A
Other languages
Chinese (zh)
Other versions
CN113158024A (en
Inventor
何向南
魏天心
冯福利
陈佳伟
易津锋
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.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
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 University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN202110218946.5A priority Critical patent/CN113158024B/en
Publication of CN113158024A publication Critical patent/CN113158024A/en
Application granted granted Critical
Publication of CN113158024B publication Critical patent/CN113158024B/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
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a causal reasoning method for correcting popularity deviation of a recommendation system, which comprises the following steps: acquiring a matching score of a user and an article in a current recommendation system; predicting an item score according to the popularity of the item, and predicting a user score according to the preference of the user; and aggregating the matching scores of the user and the articles, the article scores and the user scores, predicting the matching scores of the user and the articles, and removing the influence caused by the popularity deviation to obtain the final matching scores of the user and the articles. The method provided by the invention is a model-independent counterfactual reasoning framework, can be suitable for various recommendation systems, improves the recommendation performance of the recommendation system by eliminating the popularity deviation, and can provide better and accurate personalized recommendation content for users.

Description

Causal reasoning method for correcting popularity deviation of recommendation system
Technical Field
The invention relates to the technical field of personalized recommendation, in particular to a causal reasoning method for correcting popularity deviation of a recommendation system.
Background
Personalized recommendations have revolutionized many online applications, such as e-commerce, search engines, and conversational systems. A number of recommendation models have been developed in which the default optimization option is to reconstruct the user-item historical interactions. However, the frequency distribution of items in the interaction data is never uniform, and is affected by many factors, such as exposure mechanism, public praise effect, sales activity, product quality, and the like. In most cases, the frequency distribution of the items is long-tailed, i.e., a few popular items have a significant portion of the interactions in the data set. This biases the classic training paradigm towards recommending popular items, rather than revealing the true preferences of the user.
The overall goal of the recommendation system is to provide personalized suggestions to the user, rather than simply recommending popular items. While the conventional training paradigm, which fits user behavior data by training a recommendation model, may bias the model towards popular items. This results in a poor margarian effect, making popular items recommended more frequently and therefore more popular. Existing research attempts to solve this problem by reducing the impact of popular items on training while increasing long-tailed item weight using inverse trend weighting (IPW). Although this approach is reasonable in theory, its weighting strategy is highly sensitive, which can lead to difficulties in tuning.
Disclosure of Invention
The invention aims to provide a causal reasoning method for correcting popularity deviation of a recommendation system, which is a model-independent counterfactual reasoning framework and can be suitable for various recommendation systems.
The purpose of the invention is realized by the following technical scheme:
a causal reasoning method for correcting popularity deviations of a recommendation system comprises the following steps:
acquiring a matching score of a user and an article in a current recommendation system;
predicting an item score according to the popularity of the item, and predicting a user score according to the preference of the user;
and aggregating the matching scores of the user and the articles, the article scores and the user scores, predicting the matching scores of the user and the articles, and removing the influence caused by the popularity deviation to obtain the final matching scores of the user and the articles.
The technical scheme provided by the invention can be seen that 1) the scheme is independent of the model (namely, the recommendation system model) and can be simply applied to any recommendation system model. 2) Is insensitive to the hyper-parameters, and reduces the time and difficulty for adjusting the parameters. 3) The recommendation performance of the recommendation system can be improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is a schematic diagram of a cause and effect diagram of a recommendation system provided by an embodiment of the present invention;
FIG. 2 is a flowchart of a causal inference method for correcting popularity bias of a recommendation system according to an embodiment of the present invention;
FIG. 3 is a comparison of recommended performance on an Adressa data set according to an embodiment of the present invention;
fig. 4 is a schematic diagram illustrating an influence of the parameter c on the recommended performance according to the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention discusses the problem of popularity deviation from a novel and basic view, namely causal relation, and provides a causal reasoning method for correcting the popularity deviation of a recommendation system, wherein the causal reasoning method is a model-independent counterfactual reasoning method framework (MACR). Research has found that popularity bias exists in the direct impact from item nodes to ranking scores, so the intrinsic properties of an item are the reason for erroneously assigning them higher ranking scores. To eliminate the prejudice of popularity, it is necessary to think of the question of the counterfactual that if a model uses only the relevant information of an item, then what its ranking score will be. Therefore, a causal graph is constructed to describe important causal relationships in the recommendation process, as shown in fig. 1, the causal graph is a recommendation system causal graph, and the causal relationships between the popularity of the modeled items and the user activity and the causal relationships thereof proposed by the conventional recommendation system are described; in fig. 1, I and U are nodes in a causal graph, abstractly represent items and users of the recommendation system, directed edges of different nodes represent causal relationships between the nodes, k (knowledge) represents knowledge information extracted from integrated users and item information, and Y represents a recommendation score (i.e., a score for recommendation) finally given to the item node U corresponding to the current user node I by the recommendation system. In training, by multi-task learning, the contribution of each causal relationship to the recommendation score is calculated. During the testing process, the influence of popularity is eliminated through counterfactual reasoning. It is noted that the solution provided by the invention modifies the learning process of the recommender system, which is common to many recommender systems-it can be easily implemented in existing methods.
As shown in fig. 2, a main process of the causal inference method for correcting popularity bias of a recommendation system according to an embodiment of the present invention includes:
1. and obtaining the matching score of the user and the article in the current recommendation system.
The current recommendation system can be various existing recommendation systems, the matching score of the user and the article can directly use the ranking score in the existing recommendation system, the ranking score is obtained by taking the representing information of the article and the user as input by the recommendation system, and the calculated score can reflect the preference of the user to how much the article can be matched.
2. Predicting an item score based on the popularity of the item, and predicting a user score based on the user's preferences.
In the embodiment of the present invention, an article modeling module and a user modeling module may be provided to predict the article score and the user score, respectively, and the following description mainly applies:
an item modeling module:
Figure RE-GDA0003097323850000031
indicating the impact of the popularity of the item, with more popular items scoring higher.
A user modeling module:
Figure RE-GDA0003097323850000032
showing how much user u will interact with the item regardless of whether the user's item matches. Considering that two users are randomly recommended the same number of videos, one user may click on more videos because of broader preference or stronger activity. Such users with wide interest will get higher
Figure RE-GDA0003097323850000033
In the embodiment of the present invention, I and U each indicate a current item and user in the recommendation system, and I and U indicate that the current item and user are specific items I and users U. In the two modeling modules, the information (such as attributes or historical purchase records) of the articles and the users are used as input, and then Y is calculated through the neural network respectivelyiAnd YuThe specific calculation scheme may be implemented by conventional techniques, for example, using a single-layer fully-connected neural network for calculation.
3. And aggregating the matching scores of the user and the articles, the article scores and the user scores, predicting the matching scores of the user and the articles, and removing the influence caused by the popularity deviation to obtain the final matching scores of the user and the articles.
Aggregating the matching scores of the users and the items, the item scores and the user scores, and predicting that the matching scores of the users and the items are expressed as follows:
Figure RE-GDA0003097323850000041
wherein, the first and the second end of the pipe are connected with each other,
Figure RE-GDA0003097323850000042
representing the predicted matching score of the user u and the item i;
Figure RE-GDA0003097323850000043
represents the matching score (ranking score) of the user u with the item i in the current recommendation system,
Figure RE-GDA0003097323850000044
each represents a predicted user score and item score; sigma (.) represents a sigmoid function by which
Figure RE-GDA0003097323850000045
And
Figure RE-GDA0003097323850000046
the change is in [0,1 ]]The click probability of the range of (a) to adjust the degree of dependence on user-item matching.
The key to eliminating the prevalence bias is by eliminating the direct impact of the article on the model, and therefore, will be
Figure RE-GDA0003097323850000047
Subtracting the influence caused by the popularity deviation to obtain a final matching score:
Figure RE-GDA0003097323850000048
wherein c is a hyper-parameter, representing
Figure RE-GDA0003097323850000049
The actual state of the counter can be set according to actual conditions or experience, and the invention does not limit the specific numerical values.
Through the method, the final matching score between each user and different articles can be obtained for each user, and the recommendation list is generated based on the final matching scores of the users and the different articles and is recommended to the users; illustratively, K articles with the maximum final matching score can be selected according to the set recommended article number K, and are recommended to the user after being sorted in a descending order; for example, K may be set to 20, and the specific value of K may be set according to actual circumstances or experience, and the specific value of K is not limited in the present invention.
The scheme provided by the embodiment of the invention can realize the recommendation of unbiased by matching with multi-task learning to carry out model training and counterfactual reasoning on any existing recommendation system, and only one user modeling module Y is required to be added to the existing recommendation systemu(U) and an item modeling Module Yi(I) .1. the These modules incorporate users and items into the recommendation score, which may be implemented using a fully connected network, for example.
In the embodiment of the invention, an article modeling module, a user modeling module and a current recommendation system are used as a model for training. Applications of
Figure RE-GDA00030973238500000410
As predicted recommendation score and using loss function of recommendation system, in order to implement modeling of user and project module, a multi-task learning mode is also designed, for
Figure RE-GDA00030973238500000411
And
Figure RE-GDA00030973238500000412
an additional constraint is added, in the form of a loss function for the training phase:
L＝LO+α*LI+β*LU
wherein alpha and beta are hyper-parameters used for adjusting the balance among different tasks; l isOA loss function, L, representing the current recommendation systemILoss function, L, representing an item modeling moduleULoss function representing user modeling moduleThrough LIAnd LUAnd achieving the optimization goal of modeling the causal relationship.
The loss function of the current recommendation system is expressed as:
Figure RE-GDA0003097323850000051
the loss function of the user modeling module is represented as:
Figure RE-GDA0003097323850000052
the loss function of the item modeling module is represented as:
Figure RE-GDA0003097323850000053
wherein, (u, i) represents a user-item pair consisting of a user u and an item i, and D represents training data comprising a plurality of user-item pairs; y isuiIs the real interaction data (real matching score) in the training data, and a value of 1 represents that there is interaction between the user u and the item i, otherwise it is 0.
Training the goal to fit the true historical interaction yuiUntil a training termination condition is met (for example, the training times reach the standard, or the loss function of continuous training times does not drop, etc.); and then, calculating the final matching score of the user and the article in the manner introduced in the steps 1-3, and recommending.
In order to illustrate the effectiveness of the above-described scheme of the embodiment of the present invention, in the existing recommendation system model: experiments were performed on the matrix factorization Model (MF) and the lightweight map recommendation model (LightGCN), which are representative of the conventional recommendation system and the most advanced recommendation model, respectively, using the five real data sets shown in table 1.
Data set User' s Article with a cover Number of interactions Degree of sparseness
Adressa 13,485 744 116,321 0.011594
Globo 158,323 12,005 2,520,171 0.001326
ML10M 69,166 8,790 5,000,415 0.008225
Yelp 31,668 38,048 1,561,406 0.001300
Gowalla 29,858 40,981 1,027,370 0.000840
Table 1 data set information statistics
1. The test precision for an unbiased scene is higher: according to the method, a plurality of causal relationships are modeled in the training process by introducing causal inference, and the expressive ability of the model in an unbiased scene is improved. Table 2 shows the results of the performance comparison.
Figure RE-GDA0003097323850000061
TABLE 2 comparison of Performance of MF with MACR _ MF
The MACR _ MF in table 2 indicates that the solution provided by the present invention is applied to the MF model, i.e. the current system is the MF model as described above. R is called Recall, represents the Recall rate, and is an index for measuring the accuracy rate of recommendation; n represents NDCG (Normalized discrete cumulative gain), which is an index for measuring the sequencing effect; HR is collectively referred to as Hit Ratio and represents the Hit rate, i.e., the probability that a recommendation hits the user's interest. As can be seen from Table 2, MACR _ MF performs better than the MF model on datasets of varying data volumes and data densities.
2. The method of the present invention is model independent and can be applied to any recommendation system model, and in the above table 2, the effect of the present invention is verified on the MF model; the effect on the lightweight graph recommendation model (LightGCN) is further demonstrated below, and the effect on recommendation lists of different sizes is demonstrated, as shown in fig. 3.
In fig. 3, MACR _ LightGCN indicates that the solution provided by the present invention is applied to the LightGCN model, i.e. the current system described above is the LightGCN model. In FIG. 3, the three parts (a), (b) and (c) are the results of the verification on the indices HR @ K, NDCG @ K and Recall @ K in turn, and four lines in the three parts correspond to MACR _ LightGCN, MACR _ MF, LightGCN and MF in turn from top to bottom. It can be seen from fig. 3 that the present invention can achieve good effects when applied to different inference system models.
3. Is insensitive to the hyper-parameters, and reduces the time and difficulty for adjusting the parameters.
As shown in fig. 4, which demonstrates the effect on the results on the Adressa dataset with respect to parameter c. Compared with a LighgCN model and an MF model, the effect of the invention can be continuously improved, and the parameter is easy to adjust, and the effect can be very good by simply adjusting to about 30.
It will be understood by those skilled in the art that the value at the end of each index represents the number of recommended items for each user, for example, R @20 represents that 20 items are recommended for each user, then a Recall index is calculated (the size of the index is related to how many items are recommended), and the subsequent "K" is also similar meaning, i.e., that K items are recommended for each user, and then a Recall index is calculated.
Through the description of the above embodiments, it is clear to those skilled in the art that the above embodiments may be implemented by software, or by software plus a necessary general hardware platform. With this understanding, the technical solutions of the embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are also within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (6)

1. A causal reasoning method for correcting popularity bias of a recommendation system, comprising:
acquiring a matching score of a user and an article in a current recommendation system;
predicting item scores according to popularity of items, and predicting user scores according to preferences of users;
aggregating the matching scores of the user and the articles, the article scores and the user scores, predicting the matching scores of the user and the articles, and removing the influence caused by the popularity deviation to obtain the final matching scores of the user and the articles;
aggregating the matching scores of the user and the items, the item scores and the user scores, and predicting the matching scores of the user and the items as follows:
Figure FDA0003570183070000011
wherein the content of the first and second substances,
Figure FDA0003570183070000012
representing the predicted matching score of the user u and the item i;
Figure FDA0003570183070000013
represents the matching score of the user u and the item i in the current recommendation system,
Figure FDA0003570183070000014
each represents a predicted user score and an item score; σ (.) denotes a sigmoid function;
the calculation formula of the final matching score of the user and the article is as follows:
Figure FDA0003570183070000015
wherein c is a hyperparameter representing
Figure FDA0003570183070000016
Counter fact state of (c).
2. The causal reasoning method for correcting the popularity bias of the recommendation system of claim 1, wherein the predicting of the item score according to the popularity of the item, the predicting of the user score according to the preference of the user are respectively realized by an item modeling module and a user modeling module;
training an article modeling module, a user modeling module and a current recommendation system as a model, wherein a loss function is as follows:
L＝LO+α*LI+β*LU
wherein alpha and beta are hyper-parameters; l isOA loss function, L, representing the current recommendation systemILoss function, L, representing an item modeling moduleURepresenting the loss function of the user modeling module.
3. The causal reasoning method for correcting the popularity deviation of the recommendation system according to claim 2, wherein the penalty function of the current recommendation system is expressed as:
Figure FDA0003570183070000017
wherein, (u, i) represents a user-item pair consisting of a user u and an item i, and D represents training data comprising a plurality of user-item pairs;
Figure FDA0003570183070000018
representing the matching score of the user u and the item i in the current recommendation system, and sigma (.) representing a sigmoid function; y isuiThe data are real interactive data in the training data, the value is 1, which represents that the interaction exists between the user u and the article i, otherwise, the value is 0.
4. The causal reasoning method for correcting the recommendation system popularity bias of claim 2, wherein the loss function of the user modeling module is expressed as:
Figure FDA0003570183070000021
wherein, (u, i) represents a user-item pair consisting of a user u and an item i, and D represents training data comprising a plurality of user-item pairs;
Figure FDA0003570183070000022
representing the predicted user score, and sigma (.) representing a sigmoid function; y isuiThe data are real interactive data in the training data, the value is 1, which represents that the interaction exists between the user u and the article i, otherwise, the value is 0.
5. The causal reasoning method for correcting recommendation system popularity bias of claim 2, wherein the loss function of the item modeling module is expressed as:
Figure FDA0003570183070000023
wherein, (u, i) represents a user-item pair consisting of a user u and an item i, and D represents training data comprising a plurality of user-item pairs;
Figure FDA0003570183070000024
represents the predicted item score, σ (.) represents the sigmoid function; y isuiThe data are real interactive data in training data, the value is 1, which represents that interaction exists between the user u and the article i, otherwise, the value is 0.
6. The causal reasoning method for correcting popularity bias of recommendation systems according to claim 1, wherein recommendation lists are generated based on the size of the final matching scores of the user and different items and recommended to the user.
CN202110218946.5A 2021-02-26 2021-02-26 Causal reasoning method for correcting popularity deviation of recommendation system Active CN113158024B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110218946.5A CN113158024B (en) 2021-02-26 2021-02-26 Causal reasoning method for correcting popularity deviation of recommendation system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110218946.5A CN113158024B (en) 2021-02-26 2021-02-26 Causal reasoning method for correcting popularity deviation of recommendation system

Publications (2)

Publication Number Publication Date
CN113158024A CN113158024A (en) 2021-07-23
CN113158024B true CN113158024B (en) 2022-07-15

Family

ID=76883537

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110218946.5A Active CN113158024B (en) 2021-02-26 2021-02-26 Causal reasoning method for correcting popularity deviation of recommendation system

Country Status (1)

Country Link
CN (1) CN113158024B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114090890B (en) * 2021-11-23 2023-04-28 电子科技大学 Method for recommending anti-facts item based on graph convolution network
CN114861783B (en) * 2022-04-26 2023-05-12 北京三快在线科技有限公司 Recommendation model training method and device, electronic equipment and storage medium
CN115809374B (en) * 2023-02-13 2023-04-18 四川大学 Method, system, device and storage medium for correcting mainstream deviation of recommendation system
CN117522532B (en) * 2024-01-08 2024-04-16 浙江大学 Popularity deviation correction recommendation method and device, electronic equipment and storage medium

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103209342A (en) * 2013-04-01 2013-07-17 电子科技大学 Collaborative filtering recommendation method introducing video popularity and user interest change
CN103246672A (en) * 2012-02-09 2013-08-14 中国科学技术大学 Method and device for performing personalized recommendation on users
CN104851025A (en) * 2015-05-09 2015-08-19 湘南学院 Case-reasoning-based personalized recommendation method for E-commerce website commodity
CN110188283A (en) * 2019-06-05 2019-08-30 中国人民解放军国防科技大学 Information recommendation method and system based on joint neural network collaborative filtering
CN110197404A (en) * 2019-06-12 2019-09-03 合肥工业大学 The personalized long-tail Method of Commodity Recommendation and system of popularity deviation can be reduced
CN110781409A (en) * 2019-10-25 2020-02-11 东北大学秦皇岛分校 Article recommendation method based on collaborative filtering
CN111143702A (en) * 2019-12-16 2020-05-12 中国科学技术大学 Information recommendation method, system, device and medium based on graph convolution neural network

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102761627B (en) * 2012-06-27 2015-12-09 北京奇虎科技有限公司 Based on cloud network address recommend method and system and the relevant device of terminal access statistics

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246672A (en) * 2012-02-09 2013-08-14 中国科学技术大学 Method and device for performing personalized recommendation on users
CN103209342A (en) * 2013-04-01 2013-07-17 电子科技大学 Collaborative filtering recommendation method introducing video popularity and user interest change
CN104851025A (en) * 2015-05-09 2015-08-19 湘南学院 Case-reasoning-based personalized recommendation method for E-commerce website commodity
CN110188283A (en) * 2019-06-05 2019-08-30 中国人民解放军国防科技大学 Information recommendation method and system based on joint neural network collaborative filtering
CN110197404A (en) * 2019-06-12 2019-09-03 合肥工业大学 The personalized long-tail Method of Commodity Recommendation and system of popularity deviation can be reduced
CN110781409A (en) * 2019-10-25 2020-02-11 东北大学秦皇岛分校 Article recommendation method based on collaborative filtering
CN111143702A (en) * 2019-12-16 2020-05-12 中国科学技术大学 Information recommendation method, system, device and medium based on graph convolution neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
《UTSP: User-Based Two-Step Recommendation With Popularity Normalization Towards Diversity and Novelty》;Ke Niu，等;《IEEE Access 》;20190906;第7卷;第145426-145434页 *
基于社交信息和物品曝光度的矩阵分解推荐;韩勇，等;《浙江大学学报(工学版)》;20190131;第53卷(第1期);第89-98页 *

Also Published As

Publication number Publication date
CN113158024A (en) 2021-07-23

Similar Documents

Publication Publication Date Title
CN113158024B (en) Causal reasoning method for correcting popularity deviation of recommendation system
CN111460130B (en) Information recommendation method, device, equipment and readable storage medium
Maksai et al. Predicting online performance of news recommender systems through richer evaluation metrics
Sun et al. Learning multiple-question decision trees for cold-start recommendation
CN111582443B (en) Recommendation method based on Mask mechanism and level attention mechanism
KR20050043917A (en) Statistical personalized recommendation system
US12020267B2 (en) Method, apparatus, storage medium, and device for generating user profile
CN111242310A (en) Feature validity evaluation method and device, electronic equipment and storage medium
Zheng et al. Deviation-based contextual SLIM recommenders
Navgaran et al. Evolutionary based matrix factorization method for collaborative filtering systems
CN110222838B (en) Document sorting method and device, electronic equipment and storage medium
Salehi An effective recommendation based on user behaviour: a hybrid of sequential pattern of user and attributes of product
CN116680320A (en) Mixed matching method based on big data
CN110727872A (en) Method and device for mining ambiguous selection behavior based on implicit feedback
EP3920100A1 (en) Adaptive pointwise-pairwise learning to rank
CN113407854A (en) Application recommendation method, device and equipment and computer readable storage medium
Farahani et al. Adaptive personalized recommender system using learning automata and items clustering
CN113988951A (en) Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering
CN112991026A (en) Commodity recommendation method, system, equipment and computer readable storage medium
Wang et al. Enhancing Long Tail Recommendation Based on User's Experience Evolution
Gao et al. [Retracted] Construction of Digital Marketing Recommendation Model Based on Random Forest Algorithm
Wang et al. Multi‐feedback Pairwise Ranking via Adversarial Training for Recommender
CN115456656A (en) Method and device for predicting purchase intention of consumer, electronic equipment and storage medium
CN110956528B (en) Recommendation method and system for e-commerce platform
Sharma et al. A review on collaborative filtering using knn algorithm

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