CN109241412A - A kind of recommended method, system and electronic equipment based on network representation study - Google Patents
A kind of recommended method, system and electronic equipment based on network representation study Download PDFInfo
- Publication number
- CN109241412A CN109241412A CN201810942135.8A CN201810942135A CN109241412A CN 109241412 A CN109241412 A CN 109241412A CN 201810942135 A CN201810942135 A CN 201810942135A CN 109241412 A CN109241412 A CN 109241412A
- Authority
- CN
- China
- Prior art keywords
- user
- node
- network
- article
- occurrence
- 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.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This application involves a kind of recommended method, system and electronic equipments based on network representation study.This method comprises: step a: constructing user-article co-occurrence network based on bipartite graph network and injection shadow figure；Step b: it is directed to the user-article co-occurrence net definitions search strategy, obtains the neighbor node of each user node and article node；Step c: it according to each user node and article node and respective neighbor node, is indicated using the vector that network representation learns to obtain each user node and article node；Step d: indicating according to the vector of each user node and article node, the maximally related article node of each user node is calculated by vector, and recommend maximally related article to each user according to calculated result.The problem of the application slows down the sparsity problem of collaborative filtering, makes the explanatory stronger of recommender system, alleviates scalability in collaborative filtering significantly.
Description
Technical Field
The application belongs to the technical field of data mining and recommendation, and particularly relates to a recommendation method, a recommendation system and electronic equipment based on network representation learning.
Background
With the coming of the big data era, the recommendation system is more and more concerned, has excellent performance in helping people to quickly screen data and solve the problem of information overload, and can be used for recommending potentially favorite articles to each user in a personalized manner, such as recommendation of similar articles for Taobao, music recommendation of Internet music and the like. Today recommendation systems are widely developed and have penetrated into various aspects of people's daily life, such as: music recommendations, movie recommendations, e-commerce, cell phone applications, etc.
Along with the spread of recommendation systems, various recommendation methods have been proposed: including content-based recommendations, collaborative filtering, graph-based recommendations, etc. The matrix decomposition method for predicting the scores of the items by the users through collaborative filtering is undoubtedly one of the most successful recommendation algorithms, and the existing recommendation system mostly uses the collaborative filtering method. The collaborative filtering assumes that similar users and similar articles in the historical records are similar in the future, wherein the matrix decomposition method is applied most, the scoring information of the articles by the users is stored by using a scoring matrix and then is decomposed into a low-dimensional article matrix and a low-dimensional user matrix for multiplication, so that missing or nonexistent scoring information can be obtained, and then recommendation is completed.
The invention patent with the application number of '201410007387.3' and the name of 'collaborative filtering recommendation method based on network community' discloses a collaborative filtering recommendation method based on network community, which carries out recommendation through the following steps: acquiring scoring information of items to be recommended by a user, and indirectly generating a relationship network between the user and the user by utilizing the scoring data of the items to be recommended by the user; calculating the similarity between users; dividing the user relationship network into a plurality of user communities through community detection based on similarity; selecting k users with the maximum similarity in the community where the users are located to form a neighbor user set, and performing prediction scoring on unscored items of the target user according to the neighbor user set; recommending the largest item in the scoring predicted values to the user.
The application number of 201710799698.1 is 'resource recommendation system and method based on network learning environment' screens out a user group similar to a target user by using a collaborative filtering idea, scores and recommends learning resources by combining the similarity between the similar user group and the target user and the confidence of user scoring, so that the scoring of the learning resources has user pertinence and scoring objectivity, and personalized and high-quality learning resources are recommended for the user.
However, collaborative filtering tends to suffer from sparsity and scalability issues. First, in real life, the scoring information of the user on the articles is very little, many inactive users score few articles or many unwelcome articles, and the scoring information is concentrated in several popular articles, so the scoring matrix is sparse and irregularly distributed in practical situations. Secondly, different items of different users are often recommended by a recommendation system, so that personalized requirements are realized, however, different recommendations are performed for different users, global computation is needed in the recommendation process, and with the continuous increase of the number of the users and the number of the items, the consumption of the global computation is increased continuously, so that the expandability becomes a main problem.
Disclosure of Invention
The application provides a recommendation method, a recommendation system and electronic equipment based on network representation learning, and aims to solve at least one of the technical problems in the prior art to a certain extent.
In order to solve the above problems, the present application provides the following technical solutions:
a recommendation method based on network representation learning comprises the following steps:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step a, the constructing a user-item co-occurrence network based on a bipartite graph network and a monothetic mapping specifically includes:
step a 1: storing the scores of the user on the items by using a bipartite graph network, and constructing a user-item bipartite graph;
step a 2: storing co-occurrence relations among the articles by using the single-projection image to construct an article co-occurrence network;
step a 3: and constructing a user-item co-occurrence network based on the user-item bipartite graph and the item co-occurrence network, and setting an OT parameter and a PR parameter to filter useless co-occurrence relations in the user-item co-occurrence network.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step a3, the OT parameter is used to reflect the strength of the co-occurrence relationship between two items, and the OT calculation formula between two items is as follows:
in the above formula, the first and second carbon atoms are,for determining whether the item appears in the preference list of user i,the calculation formula of (2) is as follows:
the PR parameter is used for filtering out the influence of a user with good habits on the co-occurrence relationship of the articles, and the PR calculation formula is as follows:
in the above formula, the first and second carbon atoms are,a set of item scores representing the ith user,representing the set of favorite items of the ith user.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step b, the defined search strategy comprises a breadth-first sampling strategy and a depth-first sampling strategy, and the neighborhood nodes of the breadth-first sampling strategy are limited to the nodes directly connected to the source node; and the neighborhood nodes of the depth-first sampling strategy consist of nodes which are continuously sampled from the source node.
The technical scheme adopted by the embodiment of the application further comprises the following steps: in the step c, the obtaining of the vector representation of each user node and each article node by using network representation learning according to each user node, each article node and each neighbor node specifically includes: and (4) using skip-gram training, and combining random gradient descent and negative sampling to obtain vector representations of each user node and each article node.
Another technical scheme adopted by the embodiment of the application is as follows: a recommendation system for web-based representation learning, comprising:
the user-article co-occurrence network construction module: for constructing a user-item co-occurrence network based on the bipartite graph network and the single-shot graph;
the search strategy definition module: the system comprises a search strategy definition module, a storage module and a control module, wherein the search strategy definition module is used for defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and a neighbor node of an article node;
the network representation learning module: the system comprises a network representation learning system, a network representation learning system and a network representation learning system, wherein the network representation learning system is used for obtaining vector representations of each user node and each article node and respective neighbor nodes;
a vector calculation module: and the system is used for obtaining the most relevant article node of each user node through vector calculation according to the vector representation of each user node and article node, and recommending the most relevant article to each user according to the calculation result.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the user-item co-occurrence network construction module comprises:
bipartite graph construction unit: the user-item bipartite graph is constructed by storing user scores of items through a bipartite graph network;
an article co-occurrence network construction unit: for storing co-occurrence relationships between the items using the single-shot map, constructing an item co-occurrence network;
user-item co-occurrence network construction unit: and constructing a user-item co-occurrence network based on the user-item bipartite graph and the item co-occurrence network, and setting an OT parameter and a PR parameter to filter useless co-occurrence relations in the user-item co-occurrence network.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the OT parameter is used for reflecting the strength of the co-occurrence relation between the two articles, and the OT calculation formula between the two articles is as follows:
in the above formula, the first and second carbon atoms are,for determining whether the item appears in the preference list of user i,the calculation formula of (2) is as follows:
the PR parameter is used for filtering out the influence of a user with good habits on the co-occurrence relationship of the articles, and the PR calculation formula is as follows:
in the above formula, the first and second carbon atoms are,a set of item scores representing the ith user,representing the set of favorite items of the ith user.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the search strategy defined by the search strategy definition module comprises an breadth-first sampling strategy and a depth-first sampling strategy, and neighborhood nodes of the breadth-first sampling strategy are limited to nodes directly connected to a source node; and the neighborhood nodes of the depth-first sampling strategy consist of nodes which are continuously sampled from the source node.
The technical scheme adopted by the embodiment of the application further comprises the following steps: the network representation learning module obtains vector representations of each user node and each article node by using network representation learning according to each user node, each article node and each neighbor node specifically as follows: and (4) using skip-gram training, and combining random gradient descent and negative sampling to obtain vector representations of each user node and each article node.
The embodiment of the application adopts another technical scheme that: an electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the one processor to cause the at least one processor to perform the following operations of the above-described network representation learning-based recommendation method:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
Compared with the prior art, the embodiment of the application has the advantages that: the recommendation method, the recommendation system and the electronic equipment based on network representation learning firstly combine the bipartite graph and the single-mode projection graph to construct a user-article co-occurrence network, then set breadth-first search and depth-first search strategies to traverse network sampling, and finally obtain a recommendation result through a network representation learning method, compared with the prior art, the recommendation method and the recommendation system based on network representation learning have the following advantages:
1. the bipartite graph and the single-projection graph are combined to form a co-occurrence network to show the relationship between users and articles in the recommendation system, so that the problem of sparsity of collaborative filtering is relieved;
2. by using the depth-first and breadth-first traversal networks, the scoring relationship, the article co-occurrence relationship and the user similarity relationship of the user to the articles can be explored simultaneously, so that the interpretability of the recommendation system is stronger;
3. the network representation learning method is combined with the application of random gradient descent and negative sampling, global calculation is not needed, recommendation can be carried out, and the problem of expandability in collaborative filtering is greatly relieved.
Drawings
FIG. 1 is a flow chart of a recommendation method for web-based representation learning according to an embodiment of the present application;
FIG. 2(a) is a schematic diagram of a user-item bipartite network according to an embodiment of the present application;
FIG. 2(b) is a schematic diagram of an article co-occurrence network;
FIG. 3(a) is a schematic diagram of a user-item co-occurrence network;
FIGS. 3(b) and 3(c) are schematic diagrams of the closeness and similarity, respectively, of a user-item co-occurrence network;
FIG. 4 is a schematic diagram of a search strategy;
FIG. 5 is a schematic structural diagram of a recommendation system based on web-based representation learning according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a hardware device of a recommendation method based on network representation learning according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Please refer to fig. 1, which is a flowchart illustrating a recommendation method based on web-based representation learning according to an embodiment of the present application. The recommendation method based on network representation learning comprises the following steps:
step 100: storing the scores of the user on the items by using a bipartite graph network, and constructing a user-item bipartite graph;
in step 100, as shown in fig. 2(a), a schematic diagram of a user-article bipartite graph network according to an embodiment of the present application is shown. G ═ U, O, E, where U denotes the set of user nodes { U ═ U1,u2,u3...um}, O denotes a set of item (item) nodes { O1,o2,o3...onObviously, these two sets are disjoint. E represents the set of edges in the bipartite graph network. As shown in fig. 2(a), the user-item bipartite graph network with m-4 and n-5 has edges between the user and the item and only if the user scores the item (1-5). The weight on the side represents the score of the user on the item, and when the score is smaller than a set threshold (the threshold is set to be 3 in the application, and can be specifically set according to practical application), the user dislikes the item. Conversely, when the score is greater than the set threshold, it indicates that the user likes the item, so the side weights also reflect the user's preference for the item. Use of the present applicationSet of item scores representing the ith user, useRepresenting the set of favorite items of the ith user. For example, in FIG. 2(a)While
Step 200: storing co-occurrence relations among the articles by using the single-projection image to construct an article co-occurrence network;
in step 200, as shown in fig. 2(b), the article co-occurrence network is schematically illustrated. The present application uses a single shadowgraph to define an article co-occurrence network: and if and only if two nodes in the item set are given high scores by the same user in the user set, the two node points are connected by an edge, and the weight of the edge between the nodes represents that the two nodes appear in Co-occurrence Times (OT for short) given high scores by the same user.
Step 300: constructing a user-article co-occurrence network based on the user-article bipartite graph and the article co-occurrence network, and setting an OT parameter and a PR (Personal Rating Habit) parameter to filter useless co-occurrence relations in the user-article co-occurrence network;
in step 300, the user-article co-occurrence network not only has the interaction information between the user and the article, but also includes the co-occurrence relationship between the articles, so that the recommendation sparsity problem is alleviated. When a network is constructed, information with lower value needs to be filtered, and valuable information needs to be concentrated, so that the situation that the network density is too high is avoided. Therefore, the application filters useless information by setting two parameters of OT and PR.
Specifically, OT reflects the strength of the co-occurrence relationship between two articles, and the OT calculation formula between two articles is as follows:
in the formula (1), the first and second groups,for judging whether the article isWhether it appears in the preference list of user i,the calculation formula of (2) is as follows:
the size of the OT values of the two articles reflects the possibility that the two articles are simultaneously liked by the same user, and objectively reflects the feature similarity of the two articles. The OT values and the existing numbers thereof present long-tailed distribution, the co-occurrence relations with small OT values (such as one occurrence) are more, and in a huge network relation, the co-occurrence relations like OT being 1 can be completely ignored, because the identification degree is not available and the reference value is low, if the OT values and the co-occurrence relations are also added to the user-article co-occurrence network, the efficiency is not improved, and the calculation load of network representation learning is increased. Therefore, the co-occurrence relation that the OT value is smaller than the OT set threshold (the OT set threshold is 2 in the application, and can be specifically set according to the application) is filtered out by setting the OT parameter, and the accuracy of the recommendation result is improved.
Similarly, the application sets a PR parameter to filter out the influence of good-habit users (good-scoring rate 100%) on the co-occurrence relationship of the items:
setting the user-article co-occurrence network with OT being 2 and PR being 1, filtering out co-occurrence relations less than 2 times and simultaneously filtering out the influence of people with a good evaluation percentage on the article co-occurrence relations, and obtaining the user-article co-occurrence network schematic diagram shown in fig. 3 (a).
Step 400: defining a search strategy aiming at a user-article co-occurrence network to obtain a neighbor node sequence of each user node and each article node;
in step 400, for the user-item co-occurrence network, different search strategies are required to obtain the neighbor node sequences of each user node and each item node, and the neighbor node sequences are sent to the skip-gram for training. Before defining a search strategy, defining the compactness and the similarity in a network, wherein the compactness refers to the closeness of a low-dimensional vector obtained by learning the network representation of a node belonging to a cluster; similarity refers to the proximity to which nodes in each cluster that play the same role should be; the compactness and similarity are shown in particular in fig. 3(b) and 3 (c).
In the embodiment of the application, the defined search strategies comprise a breadth-first sampling strategy (BFS) and a depth-first sampling strategy (DFS), wherein the BFS can well find out a potential user preference structure and then find out similar users, and the DFS can find out objects and users with close relations and interaction relations among the objects, so that the scoring relation, the object co-occurrence relation and the user similarity relation of the users to the objects can be simultaneously explored, the expandability problem of a recommendation system is greatly relieved, and the interpretability of the recommendation system is enabled to be stronger.
Breadth-first sampling: the neighborhood nodes are limited to nodes that are directly connected to the source node. As shown in FIG. 4, the 3 neighbor nodes from the source node v are t, x respectively1,x2。
Depth-first sampling: the neighborhood nodes consist of nodes that are continuously sampled from the source node. In fig. 4, sampling 3 neighbor nodes starting from the source node v is: x is the number of3、x4、x5。
Step 500: according to each known user node, each article node and each neighbor node, learning by using network representation and combining random gradient descent and negative sampling to obtain vector representation of each user node and each article node;
in step 500, in the embodiment of the present application, a Skip-gram is used for training to obtain a vector representation of each user node and each item node, where the Skip-gram is a method used in natural language processing, and can predict the content of a context with a fixed size of each word and obtain a vector of each word. According to the application, the Skip-gram is applied to network representation learning, and the problem of expandability in collaborative filtering is greatly relieved by combining the application of random gradient descent and negative sampling.
Step 600: and according to the vector representation of the user nodes and the article nodes, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to the user according to the calculation result.
Please refer to fig. 5, which is a schematic structural diagram of a recommendation system based on web-based representation learning according to an embodiment of the present application. The recommendation system based on network representation learning comprises a user-article co-occurrence network construction module, a search strategy definition module, a network representation learning module and a vector calculation module.
The user-article co-occurrence network construction module: for constructing a user-item co-occurrence network based on the bipartite graph network and the single-shot graph; the user-article co-occurrence network of the embodiment of the application has the interaction information of the user and the articles, and simultaneously contains the co-occurrence relation among the articles, so that the recommendation sparsity problem is relieved.
Specifically, the user-item co-occurrence network construction module includes:
bipartite graph construction unit: the user-item bipartite graph is constructed by storing user scores of items through a bipartite graph network; fig. 2(a) is a schematic diagram of a user-item bipartite graph network according to an embodiment of the present application. G ═ U, O, E, where U denotes the set of user nodes { U ═ U1,u2,u3...um}, O denotes a set of item (item) nodes { O1,o2,o3...onObviously, these two sets are disjoint. E represents the set of edges in the bipartite graph network. As shown in fig. 2(a), the user-item bipartite graph network with m-4 and n-5 has edges between the user and the item and only if the user scores the item (1-5). The weight on the side represents the score of the user on the article, when the score is less than the set threshold (the set threshold is 3 in the application, the score can be set according to the practical application)Time, indicating that the user dislikes the item. Conversely, when the score is greater than the set threshold, it indicates that the user likes the item, so the side weights also reflect the user's preference for the item. Use of the present applicationSet of item scores representing the ith user, useRepresenting the set of favorite items of the ith user. For example, in FIG. 2(a)While
An article co-occurrence network construction unit: for storing co-occurrence relationships between the items using the single-shot map, constructing an item co-occurrence network; as shown in fig. 2(b), a schematic diagram of an article co-occurrence network is shown. The present application uses a single shadowgraph to define an article co-occurrence network: and if and only if two nodes in the item set are given high scores by the same user in the user set, the two node points are connected by an edge, and the weight of the edge between the nodes represents that the two nodes appear in Co-occurrence Times (OT for short) given high scores by the same user.
User-item co-occurrence network construction unit: the system is used for constructing a user-article co-occurrence network by combining the user-article bipartite graph and the article co-occurrence network, and setting an OT parameter and a PR parameter to filter useless co-occurrence relations in the user-article co-occurrence network; when a network is constructed, information with lower value needs to be filtered, and valuable information needs to be concentrated, so that the situation that the network density is too high is avoided. Therefore, the application filters useless information by setting two parameters of OT and PR.
Specifically, OT reflects the strength of the co-occurrence relationship between two articles, and the OT calculation formula between the two articles is as follows:
in the formula (1), the first and second groups,for determining whether the item appears in the preference list of user i,the calculation formula of (2) is as follows:
the size of the OT values of the two articles reflects the possibility that the two articles are simultaneously liked by the same user, and objectively reflects the feature similarity of the two articles. The OT values and the existing numbers thereof present long-tailed distribution, the co-occurrence relations with small OT values (such as one occurrence) are more, and in a huge network relation, the co-occurrence relations like OT being 1 can be completely ignored, because the identification degree is not available and the reference value is low, if the OT values and the co-occurrence relations are also added to the user-article co-occurrence network, the efficiency is not improved, and the calculation load of network representation learning is increased. Therefore, the co-occurrence relation that the OT value is smaller than the OT set threshold (the OT set threshold is 2 in the application, and can be specifically set according to the application) is filtered out by setting the OT parameter, and the accuracy of the recommendation result is improved.
Similarly, the application sets a PR parameter to filter out the influence of good-habit users (good-scoring rate 100%) on the co-occurrence relationship of the items:
setting the user-article co-occurrence network with OT being 2 and PR being 1, filtering out co-occurrence relations less than 2 times and simultaneously filtering out the influence of people with a good evaluation percentage on the article co-occurrence relations, and obtaining the user-article co-occurrence network schematic diagram shown in fig. 3 (a).
The search strategy definition module: the method is used for defining a search strategy aiming at the user-article co-occurrence network to obtain a neighbor node sequence of each user node and each article node; before defining a search strategy, defining the compactness and the similarity in a network, wherein the compactness refers to the closeness of a low-dimensional vector obtained by learning the network representation of a node belonging to a cluster; similarity refers to the proximity to which nodes in each cluster that play the same role should be.
In the embodiment of the application, the defined search strategies comprise a breadth-first sampling strategy (BFS) and a depth-first sampling strategy (DFS), wherein the BFS can well find out a potential user preference structure and then find out similar users, and the DFS can find out objects and users with close relations and interaction relations among the objects, so that the scoring relation, the object co-occurrence relation and the user similarity relation of the users to the objects can be simultaneously explored, the expandability problem of a recommendation system is greatly relieved, and the interpretability of the recommendation system is enabled to be stronger.
Breadth-first sampling: the neighborhood nodes are limited to nodes that are directly connected to the source node. As shown in FIG. 4, the 3 neighbor nodes from the source node v are t, x respectively1,x2。
Depth-first sampling: the neighborhood nodes consist of nodes that are continuously sampled from the source node. In fig. 4, sampling 3 neighbor nodes starting from the source node v is: x is the number of3、x4、x5。
The network representation learning module: the method comprises the steps that a network representation is used for learning according to each known user node, each known article node and each known neighbor node, and random gradient descent and negative sampling are combined to obtain vector representations of each user node and each known article node; the vector representation of each user node and each article node is obtained by using Skip-gram training, and the Skip-gram is a method used for natural language processing, can predict the content of the context with the fixed size of each word, and obtains the vector of each word. According to the application, the Skip-gram is applied to network representation learning, and the problem of expandability in collaborative filtering is greatly relieved by combining the application of random gradient descent and negative sampling.
A vector calculation module: and the method is used for obtaining the most relevant article node of each user node through vector calculation according to the vector representation of the user node and the article node, and recommending the most relevant article to the user according to the calculation result.
Fig. 6 is a schematic structural diagram of a hardware device of a recommendation method based on network representation learning according to an embodiment of the present application. As shown in fig. 6, the device includes one or more processors and memory. Taking a processor as an example, the apparatus may further include: an input system and an output system.
The processor, memory, input system, and output system may be connected by a bus or other means, as exemplified by the bus connection in fig. 6.
The memory, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules. The processor executes various functional applications and data processing of the electronic device, i.e., implements the processing method of the above-described method embodiment, by executing the non-transitory software program, instructions and modules stored in the memory.
The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data and the like. Further, the memory may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory located remotely from the processor, and these remote memories may be connected to the processing system over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input system may receive input numeric or character information and generate a signal input. The output system may include a display device such as a display screen.
The one or more modules are stored in the memory and, when executed by the one or more processors, perform the following for any of the above method embodiments:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
The product can execute the method provided by the embodiment of the application, and has the corresponding functional modules and beneficial effects of the execution method. For technical details that are not described in detail in this embodiment, reference may be made to the methods provided in the embodiments of the present application.
Embodiments of the present application provide a non-transitory (non-volatile) computer storage medium having stored thereon computer-executable instructions that may perform the following operations:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
Embodiments of the present application provide a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to perform the following:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
The recommendation method, the recommendation system and the electronic equipment based on network representation learning firstly combine the bipartite graph and the single-mode projection graph to construct a user-article co-occurrence network, then set breadth-first search and depth-first search strategies to traverse network sampling, and finally obtain a recommendation result through a network representation learning method, compared with the prior art, the recommendation method and the recommendation system based on network representation learning have the following advantages:
1. the bipartite graph and the single-projection graph are combined to form a co-occurrence network to show the relationship between users and articles in the recommendation system, so that the problem of sparsity of collaborative filtering is relieved;
2. by using the depth-first and breadth-first traversal networks, the scoring relationship, the article co-occurrence relationship and the user similarity relationship of the user to the articles can be explored simultaneously, so that the interpretability of the recommendation system is stronger;
3. the network representation learning method is combined with the application of random gradient descent and negative sampling, global calculation is not needed, recommendation can be carried out, and the problem of expandability in collaborative filtering is greatly relieved.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (11)
1. A recommendation method based on network representation learning is characterized by comprising the following steps:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
2. The recommendation method based on network representation learning according to claim 1, wherein in the step a, the constructing the user-item co-occurrence network based on the bipartite graph network and the monothetic graph specifically comprises:
step a 1: storing the scores of the user on the items by using a bipartite graph network, and constructing a user-item bipartite graph;
step a 2: storing co-occurrence relations among the articles by using the single-projection image to construct an article co-occurrence network;
step a 3: and constructing a user-item co-occurrence network based on the user-item bipartite graph and the item co-occurrence network, and setting an OT parameter and a PR parameter to filter useless co-occurrence relations in the user-item co-occurrence network.
3. The recommendation method based on network representation learning according to claim 2, wherein in said step a3, said OT parameter is used to reflect the strength of the co-occurrence relationship between two items, and the OT calculation formula between two items is:
in the above formula, the first and second carbon atoms are,for determining whether the item appears in the preference list of user i,the calculation formula of (2) is as follows:
the PR parameter is used for filtering out the influence of a user with good habits on the co-occurrence relationship of the articles, and the PR calculation formula is as follows:
in the above formula, the first and second carbon atoms are,a set of item scores representing the ith user,representing the set of favorite items of the ith user.
4. The recommendation method based on network representation learning of claim 3, wherein in the step b, the defined search strategy comprises a breadth-first sampling strategy and a depth-first sampling strategy, and the neighborhood nodes of the breadth-first sampling strategy are limited to the nodes directly connected to the source node; and the neighborhood nodes of the depth-first sampling strategy consist of nodes which are continuously sampled from the source node.
5. The recommendation method based on network representation learning according to claim 4, wherein in the step c, the obtaining of the vector representation of each user node and each article node by using network representation learning according to each user node and each article node and their respective neighbor nodes is specifically: and (4) using skip-gram training, and combining random gradient descent and negative sampling to obtain vector representations of each user node and each article node.
6. A recommendation system for web-based representation learning, comprising:
the user-article co-occurrence network construction module: for constructing a user-item co-occurrence network based on the bipartite graph network and the single-shot graph;
the search strategy definition module: the system comprises a search strategy definition module, a storage module and a control module, wherein the search strategy definition module is used for defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and a neighbor node of an article node;
the network representation learning module: the system comprises a network representation learning system, a network representation learning system and a network representation learning system, wherein the network representation learning system is used for obtaining vector representations of each user node and each article node and respective neighbor nodes;
a vector calculation module: and the system is used for obtaining the most relevant article node of each user node through vector calculation according to the vector representation of each user node and article node, and recommending the most relevant article to each user according to the calculation result.
7. The web representation learning-based recommendation system according to claim 6, wherein said user-item co-occurrence network construction module comprises:
bipartite graph construction unit: the user-item bipartite graph is constructed by storing user scores of items through a bipartite graph network;
an article co-occurrence network construction unit: for storing co-occurrence relationships between the items using the single-shot map, constructing an item co-occurrence network;
user-item co-occurrence network construction unit: and constructing a user-item co-occurrence network based on the user-item bipartite graph and the item co-occurrence network, and setting an OT parameter and a PR parameter to filter useless co-occurrence relations in the user-item co-occurrence network.
8. The recommendation system based on network representation learning of claim 7, wherein the OT parameter is used to reflect the strength of the co-occurrence relationship between two items, and the OT calculation formula between two items is:
in the above formula, the first and second carbon atoms are,for determining whether the item appears in the preference list of user i,the calculation formula of (2) is as follows:
the PR parameter is used for filtering out the influence of a user with good habits on the co-occurrence relationship of the articles, and the PR calculation formula is as follows:
in the above formula, the first and second carbon atoms are,a set of item scores representing the ith user,representing the set of favorite items of the ith user.
9. The web representation learning-based recommendation system according to claim 8, wherein the search strategies defined by the search strategy definition module include breadth-first sampling strategies and depth-first sampling strategies, the neighborhood nodes of the breadth-first sampling strategies being limited to nodes directly connected to the source node; and the neighborhood nodes of the depth-first sampling strategy consist of nodes which are continuously sampled from the source node.
10. The recommendation system based on network representation learning of claim 9, wherein the network representation learning module learns the vector representation of each user node and each article node using the network representation according to each user node and each article node and their respective neighbor nodes by: and (4) using skip-gram training, and combining random gradient descent and negative sampling to obtain vector representations of each user node and each article node.
11. An electronic device, comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the following operations of the web-representation-learning-based recommendation method of any one of items 1 to 5 above:
step a: constructing a user-item co-occurrence network based on the bipartite graph network and the single shadowgraph;
step b: defining a search strategy aiming at the user-article co-occurrence network to obtain each user node and the neighbor node of the article node;
step c: according to each user node, each article node and each neighbor node, using network representation learning to obtain vector representation of each user node and each article node;
step d: and according to the vector representation of each user node and each article node, obtaining the most relevant article node of each user node through vector calculation, and recommending the most relevant article to each user according to the calculation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810942135.8A CN109241412B (en) | 2018-08-17 | 2018-08-17 | A recommendation method, system and electronic device based on network representation learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810942135.8A CN109241412B (en) | 2018-08-17 | 2018-08-17 | A recommendation method, system and electronic device based on network representation learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109241412A true CN109241412A (en) | 2019-01-18 |
| CN109241412B CN109241412B (en) | 2020-12-11 |
Family
ID=65071536
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810942135.8A Active CN109241412B (en) | 2018-08-17 | 2018-08-17 | A recommendation method, system and electronic device based on network representation learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109241412B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918543A (en) * | 2019-02-26 | 2019-06-21 | 华中科技大学 | The link prediction method of node is directed in a kind of figure stream |
| CN110310185A (en) * | 2019-07-10 | 2019-10-08 | 云南大学 | Popular and novelty Method of Commodity Recommendation based on weighting bigraph (bipartite graph) |
| CN110335112A (en) * | 2019-06-18 | 2019-10-15 | 中国平安财产保险股份有限公司 | A kind of Products Show method, apparatus and computer storage medium |
| CN110879864A (en) * | 2019-10-29 | 2020-03-13 | 南京大学 | A Contextual Recommendation Method Based on Graph Neural Network and Attention Mechanism |
| CN111368205A (en) * | 2020-03-09 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Data recommendation method and device, computer equipment and storage medium |
| CN111797663A (en) * | 2019-08-02 | 2020-10-20 | 北京京东尚科信息技术有限公司 | Matching scene recognition method and device |
| CN112131480A (en) * | 2020-09-30 | 2020-12-25 | 中国海洋大学 | Personalized product recommendation method and system based on multi-layer heterogeneous attribute network representation learning |
| CN112464647A (en) * | 2020-11-23 | 2021-03-09 | 北京智源人工智能研究院 | Recommendation system-oriented negative sampling method and device and electronic equipment |
| CN112784171A (en) * | 2021-01-21 | 2021-05-11 | 重庆邮电大学 | Movie recommendation method based on context typicality |
| CN113255712A (en) * | 2020-02-12 | 2021-08-13 | 华为技术有限公司 | Recommendation method and device |
| CN113360764A (en) * | 2021-06-23 | 2021-09-07 | 甄付(上海)网络科技有限公司 | Social network construction method and system based on life service consumption scene |
| CN113763014A (en) * | 2021-01-05 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for determining co-occurrence relationship of items and method and device for obtaining judgment model |
| CN113888138A (en) * | 2021-10-27 | 2022-01-04 | 重庆邮电大学 | Project management method based on block chain and network representation learning recommendation |
| CN115087970A (en) * | 2020-02-12 | 2022-09-20 | 华为技术有限公司 | Recommendation system using bayesian graph convolution network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| US8660970B1 (en) * | 2009-04-23 | 2014-02-25 | The Boeing Company | Passive learning and autonomously interactive system for leveraging user knowledge in networked environments |
| CN106897254A (en) * | 2015-12-18 | 2017-06-27 | 清华大学 | A kind of network representation learning method |
| CN107391623A (en) * | 2017-07-07 | 2017-11-24 | 中国人民大学 | A kind of knowledge mapping embedding grammar for merging more background knowledges |
| CN108153912A (en) * | 2018-01-24 | 2018-06-12 | 北京理工大学 | A kind of knowledge based represents the Harmonious Matrix decomposition method of study |
-
2018
- 2018-08-17 CN CN201810942135.8A patent/CN109241412B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8660970B1 (en) * | 2009-04-23 | 2014-02-25 | The Boeing Company | Passive learning and autonomously interactive system for leveraging user knowledge in networked environments |
| CN103116639A (en) * | 2013-02-20 | 2013-05-22 | 新浪网技术(中国)有限公司 | Item recommendation method and system based on user-item bipartite model |
| CN106897254A (en) * | 2015-12-18 | 2017-06-27 | 清华大学 | A kind of network representation learning method |
| CN107391623A (en) * | 2017-07-07 | 2017-11-24 | 中国人民大学 | A kind of knowledge mapping embedding grammar for merging more background knowledges |
| CN108153912A (en) * | 2018-01-24 | 2018-06-12 | 北京理工大学 | A kind of knowledge based represents the Harmonious Matrix decomposition method of study |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918543A (en) * | 2019-02-26 | 2019-06-21 | 华中科技大学 | The link prediction method of node is directed in a kind of figure stream |
| CN110335112A (en) * | 2019-06-18 | 2019-10-15 | 中国平安财产保险股份有限公司 | A kind of Products Show method, apparatus and computer storage medium |
| CN110335112B (en) * | 2019-06-18 | 2024-07-16 | 中国平安财产保险股份有限公司 | Product recommendation method and device and computer storage medium |
| CN110310185A (en) * | 2019-07-10 | 2019-10-08 | 云南大学 | Popular and novelty Method of Commodity Recommendation based on weighting bigraph (bipartite graph) |
| CN110310185B (en) * | 2019-07-10 | 2022-02-18 | 云南大学 | Weighted bipartite graph-based popular and novel commodity recommendation method |
| CN111797663A (en) * | 2019-08-02 | 2020-10-20 | 北京京东尚科信息技术有限公司 | Matching scene recognition method and device |
| CN110879864A (en) * | 2019-10-29 | 2020-03-13 | 南京大学 | A Contextual Recommendation Method Based on Graph Neural Network and Attention Mechanism |
| CN110879864B (en) * | 2019-10-29 | 2022-06-07 | 南京大学 | Context recommendation method based on graph neural network and attention mechanism |
| CN115087970B (en) * | 2020-02-12 | 2025-03-11 | 华为技术有限公司 | Recommender Systems Using Bayesian Graph Convolutional Networks |
| CN113255712A (en) * | 2020-02-12 | 2021-08-13 | 华为技术有限公司 | Recommendation method and device |
| CN115087970A (en) * | 2020-02-12 | 2022-09-20 | 华为技术有限公司 | Recommendation system using bayesian graph convolution network |
| EP4088197A4 (en) * | 2020-02-12 | 2023-01-11 | Huawei Technologies Co., Ltd. | Recommender system using bayesian graph convolution networks |
| CN111368205A (en) * | 2020-03-09 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Data recommendation method and device, computer equipment and storage medium |
| CN112131480B (en) * | 2020-09-30 | 2022-06-10 | 中国海洋大学 | Personalized commodity recommendation method and system based on multilayer heterogeneous attribute network representation learning |
| CN112131480A (en) * | 2020-09-30 | 2020-12-25 | 中国海洋大学 | Personalized product recommendation method and system based on multi-layer heterogeneous attribute network representation learning |
| CN112464647A (en) * | 2020-11-23 | 2021-03-09 | 北京智源人工智能研究院 | Recommendation system-oriented negative sampling method and device and electronic equipment |
| CN113763014A (en) * | 2021-01-05 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for determining co-occurrence relationship of items and method and device for obtaining judgment model |
| CN112784171A (en) * | 2021-01-21 | 2021-05-11 | 重庆邮电大学 | Movie recommendation method based on context typicality |
| CN113360764A (en) * | 2021-06-23 | 2021-09-07 | 甄付(上海)网络科技有限公司 | Social network construction method and system based on life service consumption scene |
| CN113888138A (en) * | 2021-10-27 | 2022-01-04 | 重庆邮电大学 | Project management method based on block chain and network representation learning recommendation |
| CN113888138B (en) * | 2021-10-27 | 2024-05-14 | 重庆邮电大学 | Project management method based on blockchain and network representation learning recommendation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109241412B (en) | 2020-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109241412A (en) | A kind of recommended method, system and electronic equipment based on network representation study | |
| CN118467851B (en) | Artificial intelligent data searching and distributing method and system | |
| CN107944629B (en) | A recommendation method and device based on heterogeneous information network representation | |
| CN108287864B (en) | Interest group dividing method, device, medium and computing equipment | |
| US10565518B2 (en) | Collaborative feature learning from social media | |
| WO2021143267A1 (en) | Image detection-based fine-grained classification model processing method, and related devices | |
| WO2020094060A1 (en) | Recommendation method and apparatus | |
| CN111932308A (en) | Data recommendation method, device and equipment | |
| US11010411B2 (en) | System and method automatically sorting ranked items and generating a visual representation of ranked results | |
| CN112836125A (en) | A recommendation method and system based on knowledge graph and graph convolutional network | |
| CN107357793A (en) | Information recommendation method and device | |
| CN112632296A (en) | Knowledge graph-based paper recommendation method and system with interpretability and terminal | |
| CN111010592A (en) | Video recommendation method and device, electronic equipment and storage medium | |
| US20170286522A1 (en) | Data file grouping analysis | |
| CN116776004A (en) | Multi-interest sequence recommendation method, system and equipment | |
| CN110598084A (en) | Object sorting method, commodity sorting device and electronic equipment | |
| CN112464647B (en) | Recommendation system-oriented negative sampling method and device and electronic equipment | |
| CN114238766A (en) | Material recommendation method, material recommendation device and electronic equipment | |
| CN111881352A (en) | Content pushing method and device, computer equipment and storage medium | |
| CN113807926A (en) | Recommendation information generation method and device, electronic equipment and computer readable medium | |
| CN116595252A (en) | A data processing method and related device | |
| CN115062230A (en) | Data processing method, apparatus, device, medium, and program product | |
| Xu et al. | Towards annotating media contents through social diffusion analysis | |
| CN107730306A (en) | Film score in predicting and preference method of estimation based on multidimensional preference model | |
| WO2025011277A1 (en) | Method and apparatus for recommending media items, and device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |