CN111523040A - Social contact recommendation method based on heterogeneous information network - Google Patents
Social contact recommendation method based on heterogeneous information network Download PDFInfo
- Publication number
- CN111523040A CN111523040A CN202010350295.0A CN202010350295A CN111523040A CN 111523040 A CN111523040 A CN 111523040A CN 202010350295 A CN202010350295 A CN 202010350295A CN 111523040 A CN111523040 A CN 111523040A
- Authority
- CN
- China
- Prior art keywords
- path
- user
- meta
- model
- under
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000003066 decision tree Methods 0.000 claims abstract description 17
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims abstract description 14
- 239000011159 matrix material Substances 0.000 claims abstract description 14
- 238000013507 mapping Methods 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 18
- 230000003993 interaction Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 2
- 150000001875 compounds Chemical class 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000012886 linear function Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 239000013598 vector Substances 0.000 abstract description 5
- 238000007418 data mining Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
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
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a social contact recommendation method based on a heterogeneous information network, which comprises the following steps: for heterogeneous information networks, mapping functions are used
And carrying out normalization processing on the weight W of each associated side in the network, wherein the processed weight is omega. Then all path instance instances under the corresponding meta-path are searched by using a meta-path-based method, and the similarity relation s (x, y) between the objects of the same type under each meta-path is calculated. In order to deeply mine the similarity of objects, the similarity relation of the objects is projected to a Feature space Feature in a low dimension by using matrix decomposition, and each object can be characterized by a unique vector in the space. After solving the feature information under all element paths, inputting all feature data into a gradient lifting decision tree model and carrying out model trainingAnd linear relation and nonlinear relation among the features are learned, so that the accuracy of the recommended model is improved.
Description
Technical Field
The invention belongs to the technical field of data mining and information fusion of heterogeneous information, and particularly relates to a social recommendation algorithm.
Background
Traditional recommendation algorithms consist primarily of content-based recommendations, collaborative recommendations, hybrid recommendations, and model-based recommendations. Each recommendation algorithm has its own advantages and disadvantages. The collaborative filtering algorithm is an explicit data-based algorithm, and although the cold start problem of data is alleviated to some extent, the collaborative filtering algorithm still suffers from data sparsity. The content-based recommendation algorithm does not need to display scoring data, and recommends according to the historical behavior record of the user, generally recommends some articles similar to previous articles for the user, and the recommendation method easily leads to excessive personalization of the recommendation. The hybrid recommendation algorithm combines two recommendation methods, namely content-based recommendation and collaborative filtering, and effectively relieves the problems of data sparsity and cold start. The most common recommendation algorithm based on the model is recommendation based on matrix decomposition, and the algorithm projects the explicit scoring data of the user to two low-dimensional feature spaces respectively through a matrix decomposition method so as to perform effective recommendation. However, this algorithm is easily affected by the sparsity of the data and is not interpretable.
Most of the above recommendation algorithms are limited by data sparsity. The data sparsity not only enables the model to be interfered by noise data, but also influences the learning of the recommendation model, so that the recommendation model is easy to fall into an overfitting state, and the recommendation accuracy is reduced. In order to alleviate the influence caused by the data sparsity problem, some implicit feedback data such as user social information, attribute information and comment information of an article, and the like are usually introduced. The information can be used as auxiliary information of the explicit data, so that the training of the recommendation model is strengthened, and the prediction accuracy of the recommendation model is improved.
The invention mainly solves the problem of low recommendation accuracy caused by data sparsity. The idea of the invention is to introduce various heterogeneous information, such as social information of a user, interaction information of the user and an article, and attribute information of the article, to be fused into a heterogeneous information network, then to search information representation under corresponding meta-paths by designing a plurality of meta-paths and based on each meta-path, and then to project the information representation of each meta-path to a low-dimensional feature space by matrix decomposition. The features of each object correspond to a unique vector in the feature space. And finally, integrating all feature data and inputting the feature data into a gradient lifting decision tree for model training, and deeply mining the linear relation and the nonlinear relation among the features to obtain a final recommendation model. The model not only solves the problem of data sparsity, but also improves the recommendation accuracy of the model.
Disclosure of Invention
The present invention is directed to solving the above problems of the prior art. A social contact recommendation method based on a heterogeneous information network is provided. The technical scheme of the invention is as follows:
a social recommendation method based on a heterogeneous information network comprises the following steps:
in step S1, the heterogeneous information network includes various types of data, such as social information of the user, interaction information of the user with the item, and tag information of the item. In the heterogeneous information network, the weight of the associated edge is used for representing the quantization indexes of two connected objects, and for different types of association relations, a weight mapping function is adopted
Carrying out normalization processing on the weight;
in step S2, in the heterogeneous information network, the link paths between different objects result in semantic differences, such as a User-User (UU) path and a User-Item-User (umu) path, where the UU path represents a friend of the target User. And the UIU path represents a user who likes the same item as the target user. These link paths are meta-paths, which is a sequence of relationships between object types that defines a new compound relationship between the start type and the end type. Designing a plurality of meta-paths meta-path in a heterogeneous information network, traversing the heterogeneous information network based on the meta-paths, and searching all path instance instances under the meta-paths;
step S3, for all path instances under each meta-path, converting the relationship of the heterogeneous objects in the path instances into the influence relationship M (x, y) of the homogeneous objects based on the influence propagation;
step S4, under the same meta-path, summing the influence of all path instances of two similar objects, and calculating to obtain the similarity relation S (x, y) of the two objects under the meta-path;
step S5, in order to unify the information representation under the meta path, the similarity relation under the meta path is mapped to the numerical value of the [0,1] interval by using a nonlinear function;
step S6, respectively projecting the similarity data between all the objects under each element path to the corresponding low-dimensional Feature space Feature by using a matrix decomposition method;
step S7, inputting all object characteristics into a gradient lifting decision tree model, performing characteristic learning, and training a recommendation model fk(xi)；
Step S8, predicting the preference degree of the user to the articles by using the trained recommendation model, then sorting according to the preference degree from large to small, and recommending the Top-N articles to the user
Further, the step S1 adopts a weight mapping function
Carrying out normalization processing on the weight; the method comprises the following steps:
and for the sides of the user and the articles, normalizing the scores of the user by adopting a max-min method. Is given by the formula
Wherein r isuiRepresents the user u's score, r, for item iminAnd rmaxRepresenting the lowest score and the highest score of the user u; for the associated edges of the user and the social user, the similarity between the two is expressed by using the Jaccard coefficient. Is given by the formula
Wherein xuvRepresenting the degree of association between the user u and the user v, and respectively representing the number of the users of the associated user u and the number of the users of the associated user v; for the associated edges of the item and label, values of 0-1 are used for representation.
Further, in the step S2, a plurality of meta-paths, such as User-User (UU) meta-paths, are designed in the heterogeneous information network, and a search is performed from the heterogeneous information network according to a link sequence of the UU to search out all path instances under each meta-path;
further, the step S3 converts the relationship of the heterogeneous object in the path instance into the influence relationship M (x, y) of the homogeneous object based on the influence propagation for all path instance instances under each meta-path. Suppose x and y represent two users. The meta-path taken is UIU, which can be interpreted as a user who likes the same item as the target user. Searching all path instances satisfying the meta path from the heterogeneous information network according to the meta path sequence, and if one path instance is x (0.2) i (0.3) y, which indicates that the scores of the user x and the user y for the item i are 0.2 and 0.3 respectively, calculating the association degree of the user x and the user y under the path instance as M (x, y) ═ wxi·wyiWherein w isxiAnd wyiExpressed as the user x and user y scores for item i;
further, in step S4, under the same meta-path, the influence summation is performed on all path instances of two similar objects, and a similarity relation S (x, y) of the two objects under the meta-path is obtained through calculation. Is given by the formula
Wherein M measures the relevance of a path instance satisfying the meta-path, and then sums all relevance of user x and user y;
further, in step S5, in order to unify the information representation under the meta path, the similarity relationship under the meta path is mapped to [0,1] using a non-linear function]The numerical value of the interval. The mapping function is
Wherein x represents a similarity value between two objects;
further, the step S6 projects the similarity data between all the objects under each meta-path to the corresponding low-dimensional Feature space Feature by using a matrix decomposition method, specifically including:
and respectively establishing a Feature space Feature for the similarity information under each element path, and then designing a matrix decomposition model L. The model is
Where u, v represent two users respectively, Q represents a set of similarity relationships between users, suvRepresenting similarity data transformed by a mapping function, rukAnd finally, training the model by a stochastic gradient descent method, wherein the training process is a formula in which lambda is a regular coefficient and α is a learning rate, and solving to obtain all feature spaces.
Further, in step S6, all object features are input into the gradient boosting decision tree model for feature learning and training the recommendation model fk(xi) The method specifically comprises the following steps:
when the t-th base model is trained, the gradient lifting decision tree carries out model training according to the deviation between the predicted result and the real result of the first t-1 base models. The method comprises
Wherein
Denotes the predicted value of the ith instance at the t-th iteration, ft(xi) Expressed as the bias value of the ith instance at the t-th base model. Then, the t-th base model is trained through a loss function, and the model is
Wherein l is a loss function representing the difference between the predicted value and the true valueError between, omega (f)t) A regularization term representing the t-th base model. After the t model is trained, the first t basic models are connected in series to form a high-level model, recursion is carried out in sequence until the error is 0, the model training can be stopped, and finally a recommendation model formed by a plurality of decision trees connected in series is obtained
Where K denotes the number of all base models, fkRepresenting a kth basis model;
and performing predictive analysis by using the trained gradient boosting decision tree model to predict the preference degree of the user on the articles, thereby performing effective recommendation.
The invention has the following advantages and beneficial effects:
the method aims to alleviate the influence of data sparsity on a recommendation model by introducing various implicit information, form a heterogeneous information network by the various types of information, perform data mining from the heterogeneous information network, perform linear and nonlinear feature learning on the heterogeneous information by a gradient lifting decision tree, deeply mine potential information in the heterogeneous information network, and improve the prediction accuracy of the recommendation model.
In general, data mining is generally performed by using a matrix decomposition method, and explicit scoring data and implicit feedback information are projected to a low-dimensional feature space, and then a vector relationship in the feature space is used for predicting the preference degree of a user for an article. The method comprises the steps of respectively carrying out feature projection on explicit and implicit data and then training a model by utilizing the constraint relation of corresponding feature vectors. The invention firstly constructs a uniform heterogeneous information network for various heterogeneous information and then carries out deep search on the heterogeneous information network based on various meta paths. In fact, the method performs data analysis and data integration on the heterogeneous information in the heterogeneous information network to obtain more comprehensive data information. And then performing feature projection on the information of each element path by using a matrix decomposition method, and finally integrating all feature data in a gradient lifting decision tree so as to perform recommendation model training. The gradient lifting decision tree can strengthen model training, learn linear relation and nonlinear relation among feature data, and finally improve prediction accuracy of the recommended model.
Drawings
FIG. 1 is a flow chart of a social recommendation algorithm for heterogeneous information networks according to a preferred embodiment of the present invention;
table 1 is a plurality of meta-path descriptions in a heterogeneous information network;
Detailed Description
The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.
The technical scheme for solving the technical problems is as follows:
the invention relates to a social recommendation algorithm oriented to a heterogeneous information network, which fully utilizes different network element paths meta-path to deeply dig out potential value information of a user social network, item metadata and interaction data between a user and an item from the heterogeneous information network. Each meta path corresponds to unique semantic information. For heterogeneous information networks, mapping functions are used
And carrying out normalization processing on the weight W of each associated side in the network, wherein the processed weight is omega. Then all path instance instances under the corresponding meta-path are searched out based on the meta-path, and the similarity relation s (x, y) between the objects of the same type under each meta-path is calculated. In order to deeply mine the similarity of objects, the similarity relation of the objects is projected to a Feature space Feature in a low dimension by using matrix decomposition, and each object can be characterized by a unique vector in the space. After feature information under all element paths is solved, all feature data are input into the gradient lifting decision tree model and model training is carried out, and a linear relation and a nonlinear relation among features are learned, so that the accuracy of the recommendation model is improved. The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the flow chart of the implementation of the present invention includes the following steps:
step 1, searching semantic information under different network meta-paths based on the meta-paths, and performing information processing on the semantic information to convert the semantic information into a similar relation of objects of the same type.
And 2, decomposing the similarity data under each element path by using a matrix decomposition method, and projecting the similarity data to a corresponding feature space.
And 3, learning feature data under all meta-paths by using a gradient lifting decision tree model, and deeply mining the interest preference of the user and the personalized features of the articles.
The meta path information used in the present invention is shown in table 1, and the semantic information corresponding to each meta path is explained in detail.
TABLE 1
The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.
Claims (8)
1. A social recommendation method based on a heterogeneous information network is characterized by comprising the following steps:
step S1, the heterogeneous information network includes multiple types of data, including information including social information of users, interaction information of users and articles, and label information of articles, in the heterogeneous information network, the weight of the associated edge is used to represent the quantitative index of two connected objects, for different types of association, a weight mapping function is used
Carrying out normalization processing on the weight;
step S2, designing a plurality of meta-paths in the heterogeneous information network, and traversing the heterogeneous information network based on the meta-paths to search all path instances under the meta-paths, where the link paths between different objects in the heterogeneous information network cause semantic differences, such as a User-User (UU) path and a User-Item-User (umu) path, where the UU path represents a friend of a target User, and the UIU path represents a User who likes the same Item as the target User, and the link paths are meta-paths, and the meta-paths are a relationship sequence between object types and define a new compound relationship between a start type and an end type;
step S3, for all path instances under each meta-path, converting the relationship of the heterogeneous objects in the path instances into the influence relationship M (x, y) of the homogeneous objects based on the influence propagation;
step S4, under the same meta-path, summing the influence of all path instances of two similar objects, and calculating to obtain the similarity relation S (x, y) of the two objects under the meta-path;
step S5, in order to unify the information representation under the meta path, the similarity relation under the meta path is mapped to the numerical value of the [0,1] interval by using a nonlinear function;
step S6, respectively projecting the similarity data between all the objects under each element path to the corresponding low-dimensional Feature space Feature by using a matrix decomposition method;
step S7, inputting all object characteristics into a gradient lifting decision tree model, performing characteristic learning, and training a recommendation model fk(xi)；
And step S8, predicting the preference degree of the user to the articles by using the trained recommendation model, and then sequencing according to the preference degree from large to small to recommend Top-N articles to the user.
2. The method of claim 1, wherein the method is based on a heterogeneous messageThe social recommendation method of the information network is characterized in that the step S1 adopts a weight mapping function
Carrying out normalization processing on the weight; the method comprises the following steps:
for the edges of the user and the articles, the max-min method is adopted to carry out normalization processing on the scores of the user, and the formula is
Wherein r isuiRepresents the user u's score, r, for item iminAnd rmaxRepresenting the lowest score and the highest score of the user u; for the associated edges of the user and the social user, the similarity between the two is expressed by using a Jaccard coefficient, and the formula is
Wherein xuvRepresenting the degree of association between the user u and the user v, and respectively representing the number of the users of the associated user u and the number of the users of the associated user v; for the associated edges of the item and label, values of 0-1 are used for representation.
3. The method as claimed in claim 2, wherein the step S2 is designed to form a plurality of meta paths, including User-User (UU) meta paths, in the heterogeneous information network, and searches all path instances under each meta path by searching from the heterogeneous information network according to a link sequence of the UU.
4. The method according to claim 3, wherein the step S3 is implemented by converting the relationship of the heterogeneous object in the path instance into the influence relationship M (x, y) of the homogeneous object based on influence propagation for all path instance instances under each meta-path, and specifically: if x and y are represented as two users, the meta path taken is UIU, which can be interpreted as liked by the target userAnd if one path instance is x (0.2) i (0.3) y, which indicates that the scores of the user x and the user y for the item i are 0.2 and 0.3 respectively, calculating the association degree of the user x and the user y under the path instance as M (x, y) ═ wxi·wyiWherein w isxiAnd wyiRepresented as the scores for item i for user x and user y.
5. The method as claimed in claim 4, wherein in step S4, the influence of all path instances of two similar objects is summed under the same meta-path, and the similarity S (x, y) of the two objects under the meta-path is calculated, wherein the specific formula is
Where M measures the relevance of one path instance that satisfies this meta-path, and then sums all relevance of user x and user y.
6. The method for social recommendation based on heterogeneous information network as claimed in claim 5, wherein said step S5 is implemented by mapping the similarity relationship under meta path to [0,1] using non-linear function for unifying the information representation under meta path]The numerical value of the interval. The mapping function is
Where x represents the value of similarity between two objects.
7. The social recommendation method based on the heterogeneous information network according to claim 6, wherein the step S6 projects the similarity data between all the objects under each meta-path to the corresponding low-dimensional Feature space Feature by using a matrix decomposition method, which specifically includes:
similarity information under each meta path is respectivelyEstablishing a Feature space Feature, and then designing a matrix decomposition model L. The model is
Where u, v represent two users respectively, Q represents a set of similarity relationships between users, suvRepresenting similarity data transformed by a mapping function, rukFinally, training the model by a random gradient descent method, wherein the training process is a formula, lambda is a regular coefficient, α is a learning rate, and then solving to obtain all feature spaces;
8. the social recommendation method based on the heterogeneous information network as claimed in claim 7, wherein the step S6 is implemented by inputting all object features into a gradient boosting decision tree model, performing feature learning, and training a recommendation model fk(xi) The method specifically comprises the following steps:
when the t-th base model is trained, the gradient lifting decision tree carries out model training according to the deviation between the predicted result and the real result of the first t-1 base models. The method comprises
Wherein
Denotes the predicted value of the ith instance at the t-th iteration, ft(xi) Expressed as the bias value of the ith instance at the t-th base model. Then, the t-th base model is trained through a loss function, and the model is
Where l is a loss function representing the error between the predicted and true values, Ω (f)t) A regularization term representing the t-th base model. After the t model is trained, the first t basic models are connected in series to form a high-level model, recursion is carried out in sequence until the error is 0, the model training can be stopped, and finally a recommendation model formed by a plurality of decision trees connected in series is obtained
Where K denotes the number of all base models, fkRepresenting a kth basis model;
and performing predictive analysis by using the trained gradient boosting decision tree model to predict the preference degree of the user on the articles, thereby performing effective recommendation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010350295.0A CN111523040A (en) | 2020-04-28 | 2020-04-28 | Social contact recommendation method based on heterogeneous information network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010350295.0A CN111523040A (en) | 2020-04-28 | 2020-04-28 | Social contact recommendation method based on heterogeneous information network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111523040A true CN111523040A (en) | 2020-08-11 |
Family
ID=71904733
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010350295.0A Pending CN111523040A (en) | 2020-04-28 | 2020-04-28 | Social contact recommendation method based on heterogeneous information network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111523040A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112214683A (en) * | 2020-09-09 | 2021-01-12 | 华南师范大学 | Hybrid recommendation model processing method, system and medium based on heterogeneous information network |
| CN112270571A (en) * | 2020-11-03 | 2021-01-26 | 中国科学院计算技术研究所 | Meta-model training method for cold-start advertisement click rate estimation model |
| CN112925913A (en) * | 2021-03-09 | 2021-06-08 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer-readable storage medium for matching data |
| CN113836393A (en) * | 2021-08-12 | 2021-12-24 | 中国科学技术大学 | Cold start recommendation method based on preference adaptive meta-learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491540A (en) * | 2017-08-24 | 2017-12-19 | 济南浚达信息技术有限公司 | A kind of combination depth Bayesian model and the film of collaboration Heterogeneous Information insertion recommend method |
-
2020
- 2020-04-28 CN CN202010350295.0A patent/CN111523040A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107491540A (en) * | 2017-08-24 | 2017-12-19 | 济南浚达信息技术有限公司 | A kind of combination depth Bayesian model and the film of collaboration Heterogeneous Information insertion recommend method |
Non-Patent Citations (1)
| Title |
|---|
| YABIN SHAO等: "HIBoosting: A Recommender System Based on a Gradient Boosting Machine", 《IEEE ACCESS》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112214683A (en) * | 2020-09-09 | 2021-01-12 | 华南师范大学 | Hybrid recommendation model processing method, system and medium based on heterogeneous information network |
| CN112214683B (en) * | 2020-09-09 | 2024-05-14 | 华南师范大学 | Mixed recommendation model processing method, system and medium based on heterogeneous information network |
| CN112270571A (en) * | 2020-11-03 | 2021-01-26 | 中国科学院计算技术研究所 | Meta-model training method for cold-start advertisement click rate estimation model |
| CN112270571B (en) * | 2020-11-03 | 2023-06-27 | 中国科学院计算技术研究所 | Meta-model training method for cold-start advertisement click rate estimation model |
| CN112925913A (en) * | 2021-03-09 | 2021-06-08 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer-readable storage medium for matching data |
| CN112925913B (en) * | 2021-03-09 | 2023-08-29 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for matching data |
| CN113836393A (en) * | 2021-08-12 | 2021-12-24 | 中国科学技术大学 | Cold start recommendation method based on preference adaptive meta-learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111523040A (en) | Social contact recommendation method based on heterogeneous information network | |
| CN109614614B (en) | BILSTM-CRF product name identification method based on self-attention | |
| CN108647251B (en) | Recommendation sorting method based on wide-depth gate cycle combination model | |
| CN107341687B (en) | Recommendation algorithm based on multi-dimensional labels and classification sorting | |
| CN109190023A (en) | The method, apparatus and terminal device of Collaborative Recommendation | |
| CN113918832B (en) | Graph convolution collaborative filtering recommendation system based on social relationship | |
| CN113918833B (en) | Product recommendation method realized through graph convolution collaborative filtering of social network relationship | |
| CN106250545A (en) | A kind of multimedia recommendation method and system searching for content based on user | |
| CN113239264A (en) | Personalized recommendation method and system based on meta-path network representation learning | |
| CN113918834B (en) | Graph convolution collaborative filtering recommendation method fusing social relations | |
| Shi et al. | Learning from heterogeneous sources via gradient boosting consensus | |
| CN112818256B (en) | Recommendation method based on neural collaborative filtering | |
| Chen et al. | DPM-IEDA: dual probabilistic model assisted interactive estimation of distribution algorithm for personalized search | |
| CN114840745B (en) | Personalized recommendation method and system based on graph sign learning and deep semantic matching model | |
| CN113836393B (en) | Cold start recommendation method based on preference self-adaptive meta-learning | |
| CN112905906B (en) | Recommendation method and system fusing local collaboration and feature intersection | |
| CN114564594A (en) | Knowledge graph user preference entity recall method based on double-tower model | |
| CN114330291A (en) | Text recommendation system based on dual attention mechanism | |
| CN110299194A (en) | The similar case recommended method with the wide depth model of improvement is indicated based on comprehensive characteristics | |
| CN113449200B (en) | Article recommendation method and device and computer storage medium | |
| CN117370674A (en) | Multitask recommendation algorithm integrating user behaviors and knowledge patterns | |
| CN117235375A (en) | User multi-behavior recommendation method based on graphic neural network and element learning | |
| Zuo et al. | A tag-aware recommendation algorithm based on deep learning and multi-objective optimization | |
| Lin et al. | Employment Recommendation Algorithm Based on Ensemble Learning | |
| van Heerden | Self-organizing feature maps for exploratory data analysis and data mining: A practical perspective |
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 |
Application publication date: 20200811 |
|
| RJ01 | Rejection of invention patent application after publication |