CN116628341A - Recommendation method based on multi-type view knowledge comparison learning model - Google Patents
Recommendation method based on multi-type view knowledge comparison learning model Download PDFInfo
- Publication number
- CN116628341A CN116628341A CN202310716093.7A CN202310716093A CN116628341A CN 116628341 A CN116628341 A CN 116628341A CN 202310716093 A CN202310716093 A CN 202310716093A CN 116628341 A CN116628341 A CN 116628341A
- Authority
- CN
- China
- Prior art keywords
- view
- knowledge
- hierarchical
- graph
- interaction
- 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 53
- 230000003993 interaction Effects 0.000 claims abstract description 69
- 238000012512 characterization method Methods 0.000 claims abstract description 21
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000000007 visual effect Effects 0.000 claims abstract description 4
- 230000008447 perception Effects 0.000 claims description 46
- 238000010586 diagram Methods 0.000 claims description 26
- 238000013528 artificial neural network Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 230000002776 aggregation Effects 0.000 claims description 8
- 238000004220 aggregation Methods 0.000 claims description 8
- 230000002452 interceptive effect Effects 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 238000006116 polymerization reaction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000003860 storage Methods 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 241000258957 Asteroidea Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The application discloses a recommendation method based on a multi-type view knowledge comparison learning model, which comprises the steps of generating HGV, HPV and HSV by using HIE, processing the HPV and the HSV to obtain embedded representation of a set visual angle, and carrying out local comparison learning; generating an enhanced view by adopting an AUG, gathering node information through contrast learning, and updating an embedded representation of the enhanced view; acquiring structural consistency of knowledge graphs among HGV, HPV and HSV, judging whether interaction based on the knowledge graphs and the interaction graphs is damaged, and integrating the knowledge graphs and the interaction graphs to obtain contrast loss of MVKC; and optimizing MVKC according to the contrast loss, inputting the knowledge graph and the interaction graph into the optimized model, outputting the characterization of the user and the project, and obtaining the user representation, the project representation and the prediction score by summing and connecting the characterization. Compared with the traditional technology, the method can improve the recommending effect and solve the problems that knowledge graphs are scarce, long tail entities exist and the like.
Description
Technical Field
The application relates to the technical field of big data processing, in particular to a recommendation method based on a multi-type view knowledge comparison learning model.
Background
In recent years, the rapid development of network applications, especially mobile applications, enables people to conveniently browse a large amount of network information resources, and how to recommend resources (such as commodities, movies, books, etc.) meeting the requirements of users from the starfish information resources is one of the problems of researchers at present. The recommendation system (Recommendation System, RS) can effectively filter and screen information, help users search information resources meeting the requirements of users in a personalized manner, and alleviate the problem of information overload (Information Overload). The recommendation technology is continuously developed and updated, and has been widely applied to the fields of education, music, electronic commerce, social networks and the like. After collaborative filtering algorithm is proposed, the recommendation system gradually becomes a new research hotspot, and also faces the data sparseness problem (the number of the recommended items scored by the user is too small) and the cold start problem (the new recommended items and the new users have no scoring data). Deep Learning (DL) is a machine Learning algorithm with recognition, analysis and calculation, and brings new opportunities for alleviating the problems of data sparseness and cold start, and since 2015, deep Learning has been widely applied in the fields of semantic mining, face recognition, voice recognition and the like, and the gradual maturation of a Deep Learning model brings new opportunities for the development of a recommendation system. In the annual meeting of the ACM recommendation system in 2016, the focus of future research of the recommendation system is pointed out by combining the deep learning and the recommendation system, so that a great deal of research is carried out by students and research institutions at home and abroad. In 2017, personalized recommended articles related to deep learning in top-level conferences (such as ICML, NIPS, COLT and the like) in machine learning directions have increased year by year. The study considers that deep learning can automatically learn different levels of expressions and abstractions of features from data, and is an effective strategy for solving the problems of cold start, data sparseness and the like of the traditional recommendation technology.
The methods are used in the popular industry, and training models are performed by using sufficient data sets owned by the industry, so that the recommendation accuracy of a recommendation system is high. Such as the wash mall recommendation, a large amount of historical interaction data lays a solid foundation for algorithm engineers to increase accuracy.
However, in some minority group fields, a minority history interaction data field and a specific professional field, the minority group fields have a common characteristic, and the algorithm recommendation capability is not enough. The user cannot be well recommended the appropriate item. Collaborative filtering is often limited in performance due to sparsity of user-project interactions in different recommendation techniques. To alleviate sparsity, many recommendation systems have been proposed to improve recommendation quality by embedding various auxiliary information. For example, embedding scores, comments, images, which help learn various item semantics and interaction information. They are also not deficient in recommendation systems that use knowledge graphs as auxiliary information. However, when knowledge graphs are used as auxiliary information, they are not always able to fully utilize the effective information of the knowledge graphs, or to filter and reduce long tail entities and noise caused by the introduction of the knowledge graphs. The recommendation algorithm is crucial to improving the recommendation accuracy, accurate recommendation is difficult to be made for a single view due to the long tail problem, and learning recommendation in multiple views is helpful to improving the accuracy.
In view of the above, the present application proposes a recommendation method based on a multi-type view knowledge-based contrast learning model.
Disclosure of Invention
The recommendation method based on the multi-type view knowledge comparison learning model can conduct hierarchical view comparison learning, well learn the semantic coordination of users and projects, and improve the recommendation effect. The model is trained by adopting an enhanced view contrast learning method, so that the problems of scarcity of knowledge graphs, existence of long tail entities and the like can be solved.
The primary purpose of the application is to solve the technical problems, and the technical scheme of the application is as follows:
the application provides a recommendation method based on a multi-type view knowledge comparison learning model, wherein the multi-type view knowledge comparison learning model comprises a hierarchical graph comparison learning module and an enhancement graph comparison learning module, and the recommendation method comprises the following steps:
s1, generating a hierarchical structure view, a hierarchical perception view and a hierarchical collaborative view by adopting a hierarchical map contrast learning module, wherein the hierarchical structure view is formed by a variational knowledge map and an interactive mapThe combination composition is used for processing the level perception view and the level cooperative view to obtain an embedded representation of the set view angle and />
S2, embedding representation according to set visual angle and />And performing local contrast learning, and setting positive samples and negative samples of the local contrast learning.
And S3, generating an enhancement view by adopting an enhancement map contrast learning module, gathering node information from the positive sample and the negative sample through contrast learning, and updating embedded representation of the enhancement view.
S4, acquiring knowledge graph structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhanced view, judging whether interaction based on the knowledge graph and the interaction graph is damaged or not according to the size of the knowledge graph structure consistency ci, and integrating the knowledge graph and the interaction graph to obtain the comparison loss E of the multi-type view knowledge comparison learning model aug 。
S5, according to the contrast loss L aug Optimizing the multi-type view knowledge contrast learning model by utilizing a multi-task training strategy, inputting the knowledge graph and the interaction graph into the optimized model, outputting the characterization of the user and the project, and obtaining the user representation, the project representation and the prediction score by summing and connecting the characterization. .
Further, in step S1, according to different types of relationships between items, the hierarchical structure view is taken as a user-item-entity interaction view, the hierarchical perception view is taken as an item-entity interaction view, the hierarchical collaboration view is taken as a user-item interaction view, and an item embedded representation of the item-item perception view is obtained through a K nearest neighbor node algorithm, wherein the calculation mode of the item embedded representation specifically includes:
wherein S represents an item-item perception map, S ij Representing adjacency between entity i and entity j, S ij =0, then it means that there is no connection between entity i and entity j.
Further, the specific process of processing the hierarchical perception view and the hierarchical collaboration view is as follows: the hierarchical collaborative view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain a local collaborative representation of the hierarchical collaborative view, wherein the specific calculation mode is as follows:
the hierarchical perception view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain the local perception representation of the hierarchical perception view, wherein the specific calculation mode is as follows:
the aggregation method is to recursively execute K times of aggregation, wherein K represents the nearest K-term collaboration signals captured between a user and an item through a hierarchical collaborative view.
Further, the positive sample and the negative sample of the local contrast learning are specifically: selecting a random node in the current view, taking nodes with the same attribute as the random node as positive samples, and taking the rest nodes as negative samples; wherein the sources of the negative sample include: and other nodes in the current view except for the positive sample and other nodes in the view for executing the local contrast learning except for the corresponding node of the positive sample.
Further, in step S3, the specific process of generating the enhancement view by the enhancement map contrast learning module is: generating an enhanced perception diagram according to the knowledge diagram, generating an enhanced collaborative diagram according to the interaction diagram, and combining the enhanced perception diagram and the enhanced collaborative diagram to obtain a complete enhanced view; the embedded representation of the enhanced view is generated by:
wherein α (i, r, e) represents the attention correlation of entities and relationships in the polymerization process, W i,e Representing parameterized weights of the entries and entities, the ELU represents a nonlinear activation function.
Further, the embedded representation of the enhanced view is subjected to data expansion to obtain a knowledge sub-graph eta 1 (G K) and η2 (G K ) The method specifically comprises the following steps:
wherein (i, r, e) represents the knowledge graph G K Medium itemThe triplets between the connecting entities are connected, and />Expressed in terms of probability p K The mask vector is represented as a binary indicator, the mask vector being based on a bernoulli distribution.
Further, the knowledge graph structure consistency ci is specifically:
the knowledge graph structure consistency is used for evaluating the data expansibility of the interaction graph and the knowledge graph, and the higher the ci is, the less sensitive the interaction graph is to the change of topology information; the process for judging whether to destroy interaction based on the knowledge graph and the interaction graph according to the structural consistency ci of the knowledge graph specifically comprises the following steps: integrating knowledge graph G K And interaction graph G u Creating a contrast view for knowledge subgraph eta 1 (G K) and η2 (G K ) Breaking the connection between the user-item-entity corresponding to the knowledge graph structure consistency ci smaller than the set value according to the size of the knowledge graph structure consistency ci, thereby breaking the interaction graph-based sub graph phi 1 (G u) and φ2 (G u ) At this time, the L-level representation between the user and the item is stacked as:
further, the contrast loss L aug The method comprises the following steps:
wherein , and />From two hierarchical views (phi) 1 (G K ), 1 (G u)) and (φ2 (G K ), 2 (G u ) Is supervised in mutual collaboration and is associated with the user and the representation of the item +.> and />And (5) performing contrast learning.
Further, the step S5 specifically includes: by adopting a multitasking training strategy and utilizing the contrast loss L aug Optimizing the multi-type view knowledge comparison learning model, and optimizing and recommending tasks by adopting BPR loss; through inputting the knowledge graph and the interaction graph into the optimized model, cross-view comparison learning is performed through the hierarchical graph comparison learning module and the enhancement graph comparison learning module, the characterization of the user and the item is output, the characterization is summed, the item embedded representation and the enhancement view embedded representation are connected, the user representation and the item representation are output, and the item most likely to be selected by the user and the probability of clicking the item by the user are output in a prediction score mode.
The second aspect of the present application provides a recommendation system based on a multi-type view knowledge comparison learning model, comprising: the system comprises a hierarchy comparison learning module, an enhancement map comparison learning module, a model optimization module and a recommendation module.
The hierarchical contrast learning module generates a hierarchical structure view, a hierarchical perception view and a hierarchical collaborative view, wherein the hierarchical structure view is composed of a combination of a variational knowledge graph and an interactive graph, and the hierarchical sense is achievedProcessing the knowledge view and the hierarchical collaborative view to obtain an embedded representation of the set viewing angle and />Embedded representation according to the set viewing angle +.> and />And performing local contrast learning, and setting positive samples and negative samples of the local contrast learning.
The enhancement map contrast learning module generates an enhancement view, gathers node information from positive samples and negative samples through contrast learning, updates embedded representation of the enhancement view, acquires knowledge map structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhancement view, and judges whether interaction based on the knowledge map and the interaction map is damaged according to the size of the knowledge map structure consistency ci.
The model optimization module integrates the integrated knowledge graph and the interaction graph to obtain the comparison loss L of the multi-type view knowledge comparison learning model aug According to contrast loss L aug And optimizing the multi-type view knowledge contrast learning model by utilizing a multi-task training strategy.
The recommendation module inputs the knowledge graph and the interaction graph into the optimized model, outputs the characterization of the user and the project, and obtains the user representation, the project representation and the prediction score by summing and connecting the characterization.
Compared with the prior art, the technical scheme of the application has the beneficial effects that:
the application provides a recommendation method based on a multi-type view knowledge comparison learning model, which adopts a hierarchical comparison learning mode to learn the harmony of the semantics of a user and a project, can improve recommendation and effectively eliminate noise entities, acquires the comparison loss of the model to train the model by enhancing the comparison learning mode, and can well solve the problems of scarcity of a knowledge graph, existence of long-tail entities and the like.
Drawings
FIG. 1 is a flowchart of a recommendation method based on a multi-type view knowledge-versus-learning model according to the present application.
FIG. 2 is a schematic diagram of a knowledge graph processed by a multi-type view knowledge-versus-learning model in accordance with an embodiment of the application.
FIG. 3 is a schematic diagram of a recommendation system based on a multi-type view knowledge-versus-learning model according to the present application.
Detailed Description
In order that the above-recited objects, features and advantages of the present application will be more clearly understood, a more particular description of the application will be rendered by reference to the appended drawings and appended detailed description. It should be noted that, without conflict, the embodiments of the present application and features in the embodiments may be combined with each other.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, however, the present application may be practiced in other ways than those described herein, and therefore the scope of the present application is not limited to the specific embodiments disclosed below.
Example 1
As shown in fig. 1, the present application provides a recommendation method based on a multi-type view knowledge comparison learning Model (MVKC) including a hierarchical graph comparison learning module (HIE) and an enhanced graph comparison learning module (AUG), the recommendation method comprising the steps of:
s1, generating a hierarchical structure view (HGV), a Hierarchical Perception View (HPV) and a hierarchical collaborative view (HSV) by adopting a hierarchical diagram comparison learning module, wherein the hierarchical structure view consists of a combination of a variational knowledge diagram and an interactive diagram, and processing the hierarchical perception view and the hierarchical collaborative view to obtain an embedded representation of a set viewing angle and />
It should be noted that, for contrast learning, based on the obtained HSV, HPV, and HGV, a graphic encoder suitable for both views is continuously explored, and contrast learning is performed therebetween to mutually supervise local contrast learning. Wherein HSV highlights the collaboration signal between the user and the item, and is able to capture collaboration information in the collaboration view by modeling the remote connection between the user and the item interactions. From the data processing perspective, only the nearest K terms are used, the others being ignored. By doing so, each view is now obtained, namely a user-item interaction view HSV for the collaborative view, an item entity semantic view HPV for the perceptual view, and a generic user-item-entity view for the HGV.
S2, embedding representation according to set visual angle and />And performing local contrast learning, and setting positive samples and negative samples of the local contrast learning.
And S3, generating an enhancement view by adopting an enhancement map contrast learning module, gathering node information from the positive sample and the negative sample through contrast learning, and updating embedded representation of the enhancement view.
S4, acquiring knowledge graph structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhanced view, judging whether interaction based on the knowledge graph and the interaction graph is damaged or not according to the size of the knowledge graph structure consistency ci, and integrating the knowledge graph and the interaction graph to obtain the comparison loss L of the multi-type view knowledge comparison learning model aug 。
It should be noted that the enhanced view generation function of the AUG module provides strong robustness and anti-interference, uses the relational awareness knowledge embedding layer to identify heterogeneity in the knowledge graph connection structure, and uses the parameterized attention matrix to construct the entity and the context of the relationship into the designated representation, thereby avoiding tedious manual path generation.
S5, according to the contrast loss L aug Optimizing the multi-type view knowledge contrast learning model by utilizing a multi-task training strategy, inputting the knowledge graph and the interaction graph into the optimized model, outputting the characterization of the user and the project, and obtaining the user representation, the project representation and the prediction score by summing and connecting the characterization.
It should be noted that, MVKC generates multiple types of views from different perspectives, which can effectively alleviate sparsity. In order to fully learn the semantics of the knowledge graph, a hierarchical learning mode is adopted to fully learn the knowledge semantics and collaboration. Meanwhile, to reduce noise in the knowledge graph, long tail entities are filtered, and knowledge enhancement modes are used to suppress noise. By using additional supervisory information from the knowledge graph hierarchy and reinforcement to guide cross-view versus learning paradigms, user-item interactions are given more weight in gradient descent and noise is suppressed. A number of experiments performed on three common datasets not exceeding the deletion size indicate that MVKC is always superior to the current most advanced technology.
Compared with the prior art, the method and the device have the advantages that hierarchical view comparison learning is adopted, so that the harmony of the semantics of the user and the project can be well learned, and the recommendation can be improved. The problems that knowledge graphs are scarce, long tail entities exist and the like can be well solved by adopting the enhanced view contrast learning training model. By combining the two view comparison learning methods, the recommendation is further improved, so that the recommendation is very effective.
Example 2
Based on the above embodiment 1, the specific application process of the MVKC will be further described in this embodiment with reference to fig. 2.
Further, in step S1, according to different types of relationships between items, the hierarchical structure view is taken as a user-item-entity interaction view, the hierarchical perception view is taken as an item-entity interaction view, the hierarchical collaboration view is taken as a user-item interaction view, and an item embedded representation of the item-item perception view is obtained through a K nearest neighbor node algorithm, wherein the calculation mode of the item embedded representation specifically includes:
wherein S represents an item-item perception map, S ij Representing adjacency between entity i and entity j, S ij =0, then it means that there is no connection between entity i and entity j.
It should be noted that, the item-item perception map S has a relationship perception aggregation mechanism, and retains relationship information of adjacent items.
Further, the specific process of processing the hierarchical perception view and the hierarchical collaboration view is as follows: the hierarchical collaborative view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain a local collaborative representation of the hierarchical collaborative view, wherein the specific calculation mode is as follows:
the hierarchical perception view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain the local perception representation of the hierarchical perception view, wherein the specific calculation mode is as follows:
the aggregation method is to recursively execute K times of aggregation, wherein K represents the nearest K-term collaboration signals captured between a user and an item through a hierarchical collaborative view.
Further, the positive sample and the negative sample of the local contrast learning are specifically: selecting a random node in the current view, taking nodes with the same attribute as the random node as positive samples, and taking the rest nodes as negative samples; wherein the sources of the negative sample include: and other nodes in the current view except for the positive sample and other nodes in the view for executing the local contrast learning except for the corresponding node of the positive sample.
Further, in step S3, the specific process of generating the enhancement view by the enhancement map contrast learning module is: generating an enhanced perception diagram according to the knowledge diagram, generating an enhanced collaborative diagram according to the interaction diagram, and combining the enhanced perception diagram and the enhanced collaborative diagram to obtain a complete enhanced view; the embedded representation of the enhanced view is generated by:
wherein α (i, r, e) represents the attention correlation of entities and relationships in the polymerization process, W i,e Representing parameterized weights of the entries and entities, the ELU represents a nonlinear activation function.
Further, the embedded representation of the enhanced view is subjected to data expansion to obtain a knowledge sub-graph eta 1 (G K) and η2 (G K ) The method specifically comprises the following steps:
wherein (i, r, e) represents the knowledge graph G K A triplet between the item and the connecting entity, and />Expressed in terms of probability p K The mask vector is represented as a binary indicator, the mask vector being based on a bernoulli distribution.
It should be noted that the purpose of augmenting the knowledge graph is to improve the identification of items that are less affected by structural changes and more tolerant of connections to noisy entities.
Further, the knowledge graph structure consistency ci is specifically:
the knowledge graph structure consistency is used for evaluating the data expansibility of the interaction graph and the knowledge graph, and the higher the ci is, the less sensitive the interaction graph is to the change of topology information; the process for judging whether to destroy interaction based on the knowledge graph and the interaction graph according to the structural consistency ci of the knowledge graph specifically comprises the following steps: integrating knowledge graph G K And interaction graph G u Creating a contrast view for knowledge subgraph eta 1 (g K) and η2 (G K ) Breaking the connection between the user-item-entity corresponding to the knowledge graph structure consistency ci smaller than the set value according to the size of the knowledge graph structure consistency ci, thereby breaking the interaction graph-based sub graph phi 1 (G u) and φ2 (G u ) At this time, the L-level representation between the user and the item is stacked as:
it should be noted that more user-item interactions can be identified by guiding graphic contrast learning and user preferences can be represented with less bias information.
Further, the contrast loss L aug The method comprises the following steps:
wherein , and />From two hierarchical views (phi) 1 (G K ), 1 (G u)) and (φ2 (G K ), 2 (G u ) Is supervised in mutual collaboration and is associated with the user and the representation of the item +.> and />And (5) performing contrast learning.
Further, the step S5 specifically includes: by adopting a multitasking training strategy and utilizing the contrast loss L aug Optimizing the multi-type view knowledge comparison learning model, and optimizing and recommending tasks by adopting BPR loss; through inputting the knowledge graph and the interactive graph into the optimized model, cross-view comparison learning is performed through a hierarchical graph comparison learning module and an enhanced graph comparison learning module, characterization of the user and the item is output, summation is performed on the characterization, the item embedded representation and the enhanced view embedded representation are connected, the user representation and the item representation are output, and the most favorable user is output in a prediction score modeThe item that may be selected and the probability of the user clicking on the item.
It should be noted that this approach improves the predictive score of historical items compared to items without user interaction. After optimizing various types of cross-view contrast learning in the two modules, various characterizations of users and items can be obtained. By summing and connecting the above representations, we end up with the user representation and the item representation, and we can predict their scores as follows. The following effects can be obtained 1. Only the user is given, the several items that the user is most likely to select are recommended to the user. 2. And simultaneously, a user and an item are given, and the probability of clicking the item by the user is predicted.
The scheme simultaneously adopts the LIGHT-GCN and GNN clustering method to cluster the entities, thereby effectively eliminating noise entities. The artificial clustering has the effect of subjectively improving and removing noise entities, and some noise entities can be removed without any clustering method in the market. Where LIGHT-GCN is a lightweight GCN, K times are erroneously aggregated from a user, project or entity. The Light-GCN includes messaging and aggregation mechanisms, which make computing very efficient. It foregoes feature transformation and nonlinear activation of conventional graph rolling networks that are ineffective for collaborative filtering,
example 2
Based on the above embodiment 1 and embodiment 2, as shown in fig. 3, a recommendation system based on a multi-type view knowledge-versus-learning model is provided in a second aspect of the present application, which includes: the system comprises a hierarchy comparison learning module, an enhancement map comparison learning module, a model optimization module and a recommendation module.
The hierarchical comparison learning module generates a hierarchical structure view, a hierarchical perception view and a hierarchical collaboration view, wherein the hierarchical structure view consists of a combination of a variable knowledge graph and an interactive graph, and the hierarchical perception view and the hierarchical collaboration view are processed to obtain an embedded representation with a set viewing angle and />Embedded representation according to the set viewing angle +.> and />And performing local contrast learning, and setting positive samples and negative samples of the local contrast learning.
The enhancement map contrast learning module generates an enhancement view, gathers node information from positive samples and negative samples through contrast learning, updates embedded representation of the enhancement view, acquires knowledge map structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhancement view, and judges whether interaction based on the knowledge map and the interaction map is damaged according to the size of the knowledge map structure consistency ci.
The model optimization module integrates the integrated knowledge graph and the interaction graph to obtain the comparison loss L of the multi-type view knowledge comparison learning model aug According to contrast loss L aug And optimizing the multi-type view knowledge contrast learning model by utilizing a multi-task training strategy.
The recommendation module inputs the knowledge graph and the interaction graph into the optimized model, outputs the characterization of the user and the project, and obtains the user representation, the project representation and the prediction score by summing and connecting the characterization.
In a specific embodiment, a computer device is further provided, including a memory and a processor, where the memory includes a recommendation program based on a multi-type view knowledge-versus-learning model, and the recommendation program based on the multi-type view knowledge-versus-learning model implements the recommendation method based on the multi-type view knowledge-versus-learning model when executed by the processor.
In the embodiments provided herein, it should be understood that the disclosed systems and methods can be implemented in other ways. Those of ordinary skill in the art will appreciate that: all or part of the steps for implementing the above method embodiments can be implemented by hardware associated with program instructions, and the foregoing program can be stored in a computer readable storage medium, which when executed, performs steps including the above method embodiments; and the aforementioned storage medium includes: a mobile storage device, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or the like, which can store program codes.
Alternatively, the above-described embodiments of the present application can be stored in a computer-readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied essentially or in part contributing to the prior art in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device.
All or part of the methods described in the various embodiments of the application are performed. And the aforementioned storage medium includes: a removable storage device, ROM, RAM, magnetic or optical disk, or other medium capable of storing program code.
It is to be understood that the above examples of the present application are provided by way of illustration only and not by way of limitation of the embodiments of the present application. The drawings depict structural positional relationships and are merely illustrative, and are not to be construed as limiting the patent. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the application are desired to be protected by the following claims.
Claims (10)
1. The recommendation method based on the multi-type view knowledge comparison learning model is characterized in that the multi-type view knowledge comparison learning model comprises a hierarchical graph comparison learning module HIE and an enhancement graph comparison learning module AUG, and comprises the following steps:
s1, generating a hierarchical structure view, a hierarchical perception view and a hierarchical collaborative view by adopting a hierarchical diagram comparison learning module, wherein the hierarchical structure view consists of a combination of a variable knowledge diagram and an interactive diagram, and processing the hierarchical perception view and the hierarchical collaborative view to obtain an embedded representation with a set viewing angle and />
S2, embedding representation according to set visual angle and />Performing local contrast learning, and setting positive samples and negative samples of the local contrast learning;
s3, generating an enhancement view by adopting an enhancement map contrast learning module, gathering node information from a positive sample and a negative sample through contrast learning, and updating embedded representation of the enhancement view;
s4, acquiring knowledge graph structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhanced view, judging whether interaction based on the knowledge graph and the interaction graph is damaged or not according to the size of the knowledge graph structure consistency ci, and integrating the knowledge graph and the interaction graph to obtain the comparison loss L of the multi-type view knowledge comparison learning model aug ；
S5, according to the contrast loss L aug Optimizing the multi-type view knowledge contrast learning model by utilizing a multi-task training strategy, inputting the knowledge graph and the interaction graph into the optimized model, outputting the characterization of the user and the project, and obtaining the user representation, the project representation and the prediction score by summing and connecting the characterization. .
2. The recommendation method based on a multi-type view knowledge-versus-learning model according to claim 1, wherein in step S1, according to different types of relationships between items, the hierarchical structure view is taken as a user-item-entity interaction view, the hierarchical perception view is taken as an item-entity interaction view, the hierarchical collaboration view is taken as a user-item interaction view, and an item embedded representation of the item-item perception view is obtained through a K nearest neighbor node algorithm, and the calculation mode of the item embedded representation is specifically as follows:
wherein S represents an item-item perception map, S ij Representing adjacency between entity i and entity j, S ij =0, then it means that there is no connection between entity i and entity j.
3. The recommendation method based on the multi-type view knowledge-versus-learning model according to claim 2, wherein the specific process of processing the hierarchical perception view and the hierarchical collaboration view is as follows: the hierarchical collaborative view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain a local collaborative representation of the hierarchical collaborative view, wherein the specific calculation mode is as follows:
the hierarchical perception view is aggregated and transmitted through a light graph convolution neural network, and the representations output by all layers of the light graph convolution neural network are summed to obtain the local perception representation of the hierarchical perception view, wherein the specific calculation mode is as follows:
the aggregation method is to recursively execute K times of aggregation, wherein K represents the nearest K-term collaboration signals captured between a user and an item through a hierarchical collaborative view.
4. The recommendation method based on the multi-type view knowledge contrast learning model according to claim 1, wherein positive samples and negative samples of the local contrast learning are specifically: selecting a random node in the current view, taking nodes with the same attribute as the random node as positive samples, and taking the rest nodes as negative samples; wherein the sources of the negative sample include: and other nodes in the current view except for the positive sample and other nodes in the view for executing the local contrast learning except for the corresponding node of the positive sample. .
5. The recommendation method based on the multi-type view knowledge contrast learning model according to claim 1, wherein in step S3, the specific process of generating the enhancement view by the enhancement view contrast learning module is as follows: generating an enhanced perception diagram according to the knowledge diagram, generating an enhanced collaborative diagram according to the interaction diagram, and combining the enhanced perception diagram and the enhanced collaborative diagram to obtain a complete enhanced view; the embedded representation of the enhanced view is generated by:
wherein α (i, r, e) represents the attention correlation of entities and relationships in the polymerization process, W i,e Representing parameterized weights of the entries and entities, the ELU represents a nonlinear activation function.
6. The recommendation method based on the multi-type view knowledge-based contrast learning model of claim 5, wherein the embedded representation of the enhanced view is data-expanded to obtain a knowledge subgraph η 1 (G K) and η2 9G K ) The method specifically comprises the following steps:
wherein (i, r, e) represents the knowledge graph G K A triplet between the item and the connecting entity, and />Expressed in terms of probability p K The mask vector is represented as a binary indicator, the mask vector being based on a bernoulli distribution.
7. The recommendation method based on the multi-type view knowledge comparison learning model according to claim 6, wherein the knowledge graph structural consistency ci is specifically:
the knowledge graph structure consistency is used for evaluating the data expansibility of the interaction graph and the knowledge graph, and the higher the ci is, the less sensitive the interaction graph is to the change of topology information; the process for judging whether to destroy interaction based on the knowledge graph and the interaction graph according to the structural consistency ci of the knowledge graph specifically comprises the following steps: integrating knowledge graph G K And interaction graph G u Creating a contrast view for knowledge subgraph eta 1 (g K) and η2 (G K ) Breaking the connection between the user-item-entity corresponding to the knowledge graph structure consistency ci smaller than the set value according to the size of the knowledge graph structure consistency ci, thereby breaking the interaction graph-based sub graph phi 1 (G u) and φ2 (G u ) At this time, the L-level representation between the user and the item is stacked as:
8. the recommendation method based on the multi-type view knowledge-based contrast learning model according to claim 7, wherein the contrast loss L aug The method comprises the following steps:
wherein , and />From two hierarchical views (phi) 1 (G K ), 1 (G u)) and (φ2 (G K ), 2 (G u ) Is supervised in mutual collaboration and is associated with the user and the representation of the item +.> and />And (5) performing contrast learning.
9. The recommendation method based on the multi-type view knowledge-versus-learning model according to claim 8, wherein the step S5 specifically comprises: by adopting a multitasking training strategy and utilizing the contrast loss L aug Optimizing the multi-type view knowledge comparison learning model, and optimizing and recommending tasks by adopting BPR loss; through inputting the knowledge graph and the interaction graph into the optimized model, cross-view comparison learning is performed through the hierarchical graph comparison learning module and the enhancement graph comparison learning module, the characterization of the user and the item is output, the characterization is summed, the item embedded representation and the enhancement view embedded representation are connected, the user representation and the item representation are output, and the item most likely to be selected by the user and the probability of clicking the item by the user are output in a prediction score mode.
10. A recommendation system based on a multi-type view knowledge comparison learning model comprises: the system comprises a hierarchical comparison learning module, an enhancement map comparison learning module, a model optimization module and a recommendation module;
the hierarchical comparison learning module generates a hierarchical structure view, a hierarchical perception view and a hierarchical collaboration view, wherein the hierarchical structure view consists of a combination of a variable knowledge graph and an interactive graph, and the hierarchical perception view and the hierarchical collaboration view are processed to obtain an embedded representation with a set viewing angle and />Embedded representation according to the set viewing angle +.> and />Performing local contrast learning, and setting positive samples and negative samples of the local contrast learning;
the enhancement map contrast learning module generates an enhancement map, gathers node information from positive samples and negative samples through contrast learning, updates embedded representation of the enhancement map, acquires knowledge map structure consistency ci among the hierarchical structure view, the hierarchical perception view and the hierarchical collaborative view according to the embedded representation of the enhancement map, and judges whether interaction based on the knowledge map and the interaction map is damaged according to the size of the knowledge map structure consistency ci;
the model optimization module integrates the integrated knowledge graph and the interaction graph to obtain the comparison loss L of the multi-type view knowledge comparison learning model aug According to contrast loss L aug Optimizing a multi-type view knowledge comparison learning model by utilizing a multi-task training strategy;
the recommendation module inputs the knowledge graph and the interaction graph into the optimized model, outputs the characterization of the user and the project, and obtains the user representation, the project representation and the prediction score by summing and connecting the characterization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310716093.7A CN116628341A (en) | 2023-06-15 | 2023-06-15 | Recommendation method based on multi-type view knowledge comparison learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310716093.7A CN116628341A (en) | 2023-06-15 | 2023-06-15 | Recommendation method based on multi-type view knowledge comparison learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116628341A true CN116628341A (en) | 2023-08-22 |
Family
ID=87609896
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310716093.7A Pending CN116628341A (en) | 2023-06-15 | 2023-06-15 | Recommendation method based on multi-type view knowledge comparison learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116628341A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117131282A (en) * | 2023-10-26 | 2023-11-28 | 江西财经大学 | Multi-view graph contrast learning recommendation method and system integrating layer attention mechanism |
-
2023
- 2023-06-15 CN CN202310716093.7A patent/CN116628341A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117131282A (en) * | 2023-10-26 | 2023-11-28 | 江西财经大学 | Multi-view graph contrast learning recommendation method and system integrating layer attention mechanism |
| CN117131282B (en) * | 2023-10-26 | 2024-01-05 | 江西财经大学 | Multi-view graph contrast learning recommendation method and system integrating layer attention mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112395442B (en) | Automatic identification and content filtering method for popular pictures on mobile internet | |
| CN111709518A (en) | Method for enhancing network representation learning based on community perception and relationship attention | |
| CN109947987B (en) | Cross collaborative filtering recommendation method | |
| CN112364242B (en) | Graph convolution recommendation system for context awareness | |
| CN113918834B (en) | Graph convolution collaborative filtering recommendation method fusing social relations | |
| CN114238750A (en) | Interactive visual recommendation method based on heterogeneous network information embedding model | |
| CN116628341A (en) | Recommendation method based on multi-type view knowledge comparison learning model | |
| CN116401380B (en) | Heterogeneous knowledge graph-oriented contrast learning prediction method and system | |
| Alhamdani et al. | Recommender system for global terrorist database based on deep learning | |
| Liu et al. | Research and citation analysis of data mining technology based on Bayes algorithm | |
| Liu et al. | A conflict evidence fusion method based on the composite discount factor and the game theory | |
| WO2022206103A1 (en) | User portrait construction method based on microblog heterogeneous information | |
| Xu et al. | TDNet: transformer-based network for point cloud denoising | |
| CN113254580B (en) | Special group searching method and system | |
| CN114637857A (en) | Knowledge graph convolutional network recommendation method based on denoising | |
| WO2020147259A1 (en) | User portait method and apparatus, readable storage medium, and terminal device | |
| Rao et al. | Imbalanced customer churn classification using a new multi-strategy collaborative processing method | |
| CN113343123A (en) | Training method and detection method for generating confrontation multiple relation graph network | |
| CN116861923A (en) | Multi-view unsupervised graph contrast learning model construction method, system, computer, storage medium and application | |
| CN116633589A (en) | Malicious account detection method, device and storage medium in social network | |
| CN116257696A (en) | Service recommendation method and system based on cross-modal knowledge graph comparison learning | |
| TWM583089U (en) | Smart credit risk assessment system | |
| CN115344794A (en) | Scenic spot recommendation method based on knowledge map semantic embedding | |
| CN113744023A (en) | Dual-channel collaborative filtering recommendation method based on graph convolution network | |
| Quan | Visualization and Analysis Model of Industrial Economy Status and Development Based on Knowledge Graph and Deep Neural Network |
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 |